Home , Piczo

Gesellschaft für Informatik (GI) GI-Edition publishes this series in order to make available to a broad public recent findings in informatics (i.e. computer science and informa- Lecture Notes tion systems), to document conferences that are organized in cooperation with GI and to publish the annual GI Award dissertation. in Informatics

Broken down into the fields of • Seminar • Proceedings • Dissertations • Thematics Christian Erfurth, Gerald Eichler, current topics are dealt with from the fields of research and Volkmar Schau (Eds.) development, teaching and further training in theory and practice. The Editorial Committee uses an intensive review process in order to ensure the high level of the contributions. th The volumes are published in German or English. 9 International Conference CS 2009 2 Information: http://www.gi-ev.de/service/publikationen/lni/ on Innovative Internet Community Systems Schau (Eds.): I

. 2

,V I CS 2009 Eichler G. h, rt June 15 – 17, 2009 Jena, Germany C. Erfu ISSN 1617-5468 ISBN 978-3-88579-242-0

"Tradition meets Innovation" – that is the promising motto of the 9th annual Interna- tional Conference on Innovative Internet Community Systems (I2CS) 2009, focusing on foundations,technology, applications and socialization around modern com- 148 Proceedings munity systems.Due to the rapid evolution of Web 2.0 technologies and rich mobile devices,ICT support leads to creative solutions and multi-disciplinary collaboration among researchers and industry partners.Bringing them together, this conference offers a great podium to discuss state-of-the-art,demonstrate scientific results and have a look into the future of interaction.This volume publishes contributions from the refereed program.

Christian Erfurth, GeraldEichler,Volkmar Schau(Eds.)

9th International Conference on

InnovativeInternet CommunitySystems I2CS 2009

June 15 –17, 2009 Jena, Germany

Gesellschaft für Informatik e.V. (GI) Lecture Notes in Informatics(LNI) -Proceedings Series of theGesellschaft fürInformatik(GI)

Volume P-148

ISBN 978-3-88579-242-0 ISSN 1617-5468

Volume Editors Dr.-Ing.ChristianErfurth Friedrich-Schiller-UniversitätJena, Institut fürInformatik Ernst-Abbe-Platz 2, 07743 Jena, Germany Email: [email protected] Dipl.-Ing.GeraldEichler Deutsche TelekomAG, Deutsche TelekomLaboratories, InnovationDevelopment Deutsche-Telekom-Allee7,64295 Darmstadt,Germany Email: [email protected] Dipl.-Inf.Volkmar Schau Friedrich-Schiller-UniversitätJena, Institut fürInformatik Ernst-Abbe-Platz 2, 07743 Jena, Germany Email: [email protected]

SeriesEditorial Board HeinrichC.Mayr, Universität Klagenfurt, Austria (Chairman, [email protected]) Hinrich Bonin, Leuphana-UniversitätLüneburg,Germany DieterFellner, Technische Universität Darmstadt, Germany UlrichFlegel, SAPResearch,Germany UlrichFrank, Universität Duisburg-Essen, Germany Johann-Christoph Freytag,Humboldt-Universität Berlin,Germany UlrichFurbach,Universität Koblenz, Germany Michael Goedicke, UniversitätDuisburg-Essen Ralf Hofestädt, Universität Bielefeld Michael Koch,Universität der Bundeswehr,München,Germany Axel Lehmann, Universitätder Bundeswehr München, Germany Ernst W. Mayr,Technische UniversitätMünchen, Germany SigridSchubert, Universität Siegen,Germany Martin Warnke,Leuphana-UniversitätLüneburg, Germany

Dissertations Dorothea Wagner, UniversitätKarlsruhe,Germany Seminars Reinhard Wilhelm, Universitätdes Saarlandes, Germany Thematics Andreas Oberweis,UniversitätKarlsruhe (TH)

With thewideuse of theInternet, it becomesmoreand more clear that an efficient global useofresourcesrequires ajoint discussion of problemsofcomputer networks,contents anduserbehavior. This needs alsothe applicationofmoderntheoretical approaches foranalysisand synthesisofefficient service structures.Inthiscontext, researchers from theuniversities of Ilmenau, Rostock andLeipzig started investigations under thetitle "Context-basedsearch in large decentralized networks"which finallyresulted in are- search project founded by theGermanResearch Association(DFG).InJune2001, Tho- masBöhmeand Herwig Unger organized theveryfirst internationalworkshopon"In- novative Internet ComputingSystems (IICS)", with this motivation,justasasmall meeting to presenttheir ideas andthe firstresearchresults within abiggerauditoryin Ilmenau(Thuringia). Consequently,the main topics were from thearea of computer networks,textprocessing anddataminingaswellastheoretical contributions –the main research areas of theparticipating teams from thesethree locations.

After twomoreconferences had been organized in theframe of theproject cooperation in GermanynamelyinKühlungsborn (2002) andLeipzig(2003), theyear 2004 wasa milestoneinthe historyofthe I2CS conference becausetwo majorchanges appeared. Following atrendinresearch, thenameofthe conferencewas changed into "Innovative Internet Community Systems" –justaslight changebut representing thehugedevelop- ment in theInternet area, to consider theupcomingsocialnetworks andcommunity systems.Inaddition, this conference wasthe firstone held abroad in Guadalajara (Mex- ico) to underlinethe internationalimportance andworld wide timelinesofthe topics addressed.Due to alot of colleagueswhich visit theI2CS regularly andvolunteeringfor this event, in 2005 to 2008 theconference wasorganized annually in Paris(France), Neuchâtel (Switzerland), Munich (Germany)and Martinique(France).

Of course, thebroad interest in theInternet andthe World Wide Webgenerated alot of similar conferences within theseyears anditbecomesmoreand more difficulttoattract newparticipants.Due to alot of colleagues feelingastrong tie to ourevent,I2CS could survivethisdifficult time.Fromnow on,alternatelyaninternationaland aGermanloca- tion will be selected by thesteeringcommittee to arrange the three-daysevent whichhas reached thestatusofaninternational conference.

TheInnovative Internet CommunitySystems conference in 2009 is quite anothermile- stone: notonlybecauseofthe conference returned to Germany once more. With the engagement forour yearlyevent by GeraldEichler, ChristianErfurth andVolkmar Schau,traditionallyinthe 3rd week of June,anew era of I2CS hasbeen started. In the program of the2009th conference youwillfindcompletelyrenewed topics,acommit- ment to areliable publisher as well as many more peopleinvolvedinthe organizationof theevent.All this waspossible sincethe organizers started anew,not only financial cooperationwiththe German SocietyofInformatics (GI) as well as with Deutsche Tele- komLaboratories,Germany,which hasbeen so successfulthatwehope forits conse- quentcontinuationwithinthe next years. However, with alimitationtoabout50participants andthe continuedhardpeer review selectionofpapers, I2CS staysasmall conference where personal contactsare easily possible in aquiet inspiringatmosphere, discussionand brainstorming meetingsinsmall groups are encouragedand significantly contributetothe successand special flair of the conference. Of course, we hope to continue this traditionwithour next,the 10th I2CS conference, in Bangkok(Thailand)in2010.

With kind regards on behalfofthe wholeI2CS steeringcommittee

HerwigUnger,FernuniversitätHagen,Germany Thomas Böhme, Technische UniversitätIlmenau, Germany Preface –AShortGuide to I2CS 2009

"TraditionmeetsInnovation" –thatisthe promising motto of the9th InternationalConfe- renceonInnovativeInternetCommunity Systems(I2CS). Topics,beyondthe state of the art, are discussed in an atmosphere of grandtradition. In Jena,located in Thuringia–the green heartofGermany –atthe banksofthe riverSaale, Johann Wolfgang vonGoethe, Friedrich Schiller, Georg WilhelmFriedrich Hegel, andJohannGottlieb Fichte left their mark on intellectuallife. Ernst Abbe,Carl Zeiss, andOtto Schott laid thefoundations for economic prosperity in theoptical industries.Germany's "Science City 2008" with its 450 years old Friedrich SchillerUniversitywill provide an excellent venuefor theI2CS's conference andworkshoptopics.

In thecontext of theI2CS,weare seekingfor innovationinthe area of communitysystems.The rapidevolutionofweb technologiesand rich mobile devices providethe chance to enhanceICT supportfor communities on thenextquality level. Tryingto achieve this,weprovokenew researchquestions in awiderange of connected fields. Besides emergingtechnologies,social aspects come into spotlight toofor amoreconve- nientdaily life of communities. In searchofinnovative solutions,multi-disciplinary collaboration amongresearchers andindustrypartners is essential. Hence, thegoalof this conference is to bringtogether researchers, experts, andpractitioners from various areas relatedtonovel Internet Community Systems, enhancingthe Webx.0 paradigms.

TheselectionofI2CS topics 2009 encompasses awiderange of aspects, bundled into the areas:foundations,technology, applications as well as socializations.

Foundations –Theories, models, algorithms forcommunities ! Distributed algorithms andsimulationmodels ! Game theory, graph theoryand costmodels ! Innovativecommunicationprotocols ! Self organizationand selfstabilization ! Security andprivacy protection ! Interoperability andIT-governance

Technology –Distributed architecturesand frameworks ! Service-orientedarchitectures forcommunities ! Peer-to-peer andgridarchitectures ! Distributed community middleware forWeb x.0 ! Software agents forcommunity support ! Adaptivecooperativeinformation systems ! Community management in ad-hocenvironments ! Informationretrieval anddistributed ontologies Applications andsocialization–Communities on themove ! Mobile Internet applications'experiences ! Contextand locationawareness ! Personalizationofcomponents andtools ! Personal social networks anduserbehavior ! Socialand business aspects of user generated content ! Expert profiles, collaborativefilteringand matching ! Domain specific languagesfor semantic design

Theconference track is composed of onekeynote, twoinvited talks,one panel, awork- shop, dedicated to mobile agenttechnologiesand sixpresentationsessions.The sessions are focusing on thefollowing subjects:

! Knowledgeand ContentManagement(pp. 13) ! CooperativeInformationSystems (pp. 47) ! Communitiesonthe Move(pp. 83) ! Webportals andUsability (pp. 121) ! Graph Theory,Routingand Layering (pp. 147) ! Semantic Web Technologies (pp. 181)

To ensure ahighquality of contributions,all proposed papersweresubject to twoupto four reviews by members of theprogramcommittee. Some originallyweakpapers expe- rienced an intensiveiterative improvement process. Only accepted papers are part of theseconferenceproceedings,following theorder of theconferenceprogram. We would liketothank themembers of theprogram committee andadditionalreviewers fortheir patient andflexiblesupport.

We wish theconferenceagreat success. For furtherdetails, pleaseexplore theconference websiteatthe URLhttp://i2cs.uni-jena.de/. Welcome to Jena.Enjoy your stay!

YoursI2CS 2009 organizingteam

Gerald Eichler,DeutscheTelekom Laboratories, Darmstadt/Berlin,Germany ChristianErfurth,Friedrich-Schiller-UniversitätJena/GI e.V., Germany VolkmarSchau,Friedrich-Schiller-UniversitätJena, Germany Program Committee

S. Albayrak,TUBerlin,Germany H. Arnold,DeutscheTelekom Laboratories,Germany G. Babin, HECMontreal,Canada A. Böhm,T-Systems,Germany T. Böhme, TU Ilmenau, Germany D. Chase, T-Mobile International, United Kingdom G. Eichler, Deutsche TelekomLaboratories,Germany C. Erfurth,Friedrich Schiller UniversityJena, Germany H. Fouchal, University of Antilles-Guyane, France W. Halang, FU Hagen/GI, Germany G. Heyer, LeipzigUniversity,Germany H. Höpfner, InternationalUniversityBruchsal/GI, Germany P. Hunel, University of Antilles-Guyane, France J. Kacprzyk,PolishAcademy of Science, Poland V. Kirova,Alcatel-Lucent,U.S.A. P. Kropf,UniversityofNeuchâtel, Switzerland K. Kyandoghere, University of Klagenfurt,Austria U. Lechner, University of BundeswehrMünchen,Germany F. Lehner, PassauUniversity /GI, Germany P. Meesad,KingMongkut'sUniversityofTechnology NorthBangkok, Thailand A. Mikler, University of Northern Texas, U.S.A. C. Prehofer, Nokia ResearchCenter, Finland L. Rokach,Ben-Gurion University,Israel W. Rossak, FriedrichSchillerUniversityJena, Germany H. Sack,HPI,University Potsdam, Germany V. Schau,Friedrich Schiller UniversityJena, Germany H. Schilder, nexumAG/GI, Germany H. Unger, FU Hagen, Germany K. S. Tang, City UniversityHongKong,Hong Kong M. Welsch,IBM,Germany L. Wienhofen, SINTEF,Norway

Additional Reviewers

B. Berde K.-H. Lüke P.-C.Chen M. Moganti P. Kapauan M. Will Organizing Committee

C. Erfurth,Friedrich Schiller UniversityJena, Germany (ConferenceChair) G. Eichler, DeutscheTelekom Laboratories,Germany(Conference Chair) V. Schau,Friedrich Schiller UniversityJena, Germany (Program Chair) Sponsors

Organized by

Supported by

GESELLSCHAFTFÜR INFORMATIKE.V. RegionalgruppeOstthüringen/Jena&Fachgruppe MMS Table of Contents

Session1.Knowledgeand Content Management 13

Personal KnowledgeManagement: TheroleofWeb 2.0Tools forManaging of KnowledgeatIndividualand CollectiveLevel 15 KathrinKirchner,Liana Razmerita,Thierry Nabeth Enabling Social Media content qualityassurance usingSNA 27 Maria-AmparoSanmateu, Matthias Trier,Andreas Rederer,Andreas Lienicke Contributing andsocialization-biaxialsegmentationfor usersgenerating content 36 Hendrik Send, Daniel Michelis

Session 2. Cooperative Information Systems 47

Enhancinggroup communicationusing an IMS-basedcommunity based infrastructure49 Zuzana KrifkaDobes,Karl-HeinzLüke, Andreas Rederer TowardsInternet CommunitiestoHelp Improvethe Wellbeingand RehabilitationofClinicallyStableChronic Patients 60 LeendertWienhofen, Ingrid Svagård Bi-directionalDistributionofeLearningContent forCross-technology LearningCommunities70 Raphael Zender,EnricoDressler,UlrikeLucke, DjamshidTavangarian

Session 3. Communitiesonthe Move 83

TheDevelopment of aPersonal MobileGIS 85 Ka-ho Ng,Wallace Tang FacebookAgent-an Agent-Enhanced Social (Mobile)Network Application97 Volkmar Schau,ChristianErfurth,René Pasold, WilhelmRossak Integrated Solutions andServices in Public TransportonMobile Devices 109 Karl-Heinz Lüke, Holger Mügge, MatthiasEisemann, Anke Telschow

Session 4. WebPortals andUsability 121

AHybridApproach to Identifying User Interests in Web Portals123 Fedor Bakalov, Birgitta König-Ries, AndreasNauerz, MartinWelsch Web 2.0asanautopoietic system -implications forinnovative web-interfaces -135 KathrinVent Session 5. Graph Theory,Routing and Layering 147

Agametheoretic approach to graph problems149 Thomas Böhme, Jens Schreyer SplittingOverlayNetwork forPeer-to-Peer-based MassivelyMultiplayer Online Games157 Cheng Liu, Wentong Cai An AdaptivePolicyRouting with ThermalField Approach 169 Lada-OnLertsuwanakul, HerwigUnger

Session 6. Semantic WebTechnologies 181

FromCommunitytowards Enterprise-ataxonomy-based search forexperts 183 GeraldEichler,Andreas Lommatzsch, Thomas Strecker,DanutaPloch, Conny Strecker,RobertWetzker RecommendingRelated ArticlesinWikipediavia aTopic-BasedModel 194 Wongkot Sriurai, PhayungMeesad, ChoochartHaruechaiyasak Linkingtele-TASKvideo portaltothe Semantic Web204 Harald Sack, BertBaumann,Andreas Groß, Christoph Meinel

Index of Authors 217 Session1

Knowledgeand ContentManagement

Personaland CollectiveKnowledge Management in the Web2.0: TwoFaces of Knowledge Management?

_;NqRpO _pRvqOtR1l^p;O; ];*PtRpN;mlEqptRR+ Z;8tNqj

1&RptuRpvqk\vqpTTtRkCOp2tRQpN+ btO;l%tRP;O+ W;NqRpOhWpRvqOtRK5OpkYtO;hut mF3XtOq;rtO G5QpOtQQ\vq33Tl )tOP;RW Tp;O;hR;*PtRpN;Kv8QhuW j#Z\(H)l &3ON;pOt8Tt;5l &R;Ovt NqptRR+hO;8tNqKpOQt;uhtu5

Abstract: At8 mh4 q;Q RtpO2tONtu Nqt v3OvtXN 3s _O3.Tturt [;O;rtPtON N3.;RuQ;2pQp3O ;pPpOr ;N s;vpTpN;NpOr pONtR;vNp3Ol v33XtR;Np3O ;Ou WO3.Tturt t-k vq;Ort 3s pOup2pu5;TQl rR35XQ ;Ou v3PP5OpNptQhEqpQ ;RNpvTt NRptQ N3 ;OQ.tR N.3 V5tQNp3OQJ F;O At8 mh4 N33TQ 8t ;XXTptu s3R P;O;rpOr WO3.Tturt ;N Nqt pOup2pu5;T ;Ou v3TTtvNp2tTt2tT' ;Ou $3. tsstvNp2t ;Rt At8 mh4 N33TQs3R Q5XX3RNpOr atRQ3O;T _O3.Tturt[;O;rtPtON'HsNtR pONR3u5vpOrNqt v3OvtXN 3s atRQ3O;T _O3.Tturt [;O;rtPtON 7a_[6l F3TTtvNp2t[;O;rtPtON 7F_[6 ;Ou At8 mh4 XqtO3PtO3Ol NqpQ X;XtRQ5R2t+Q upsstRtON At8 mh4 N33TQ;Ou v3PX;Rt NqtpR R3Tt s3R Q5XX3RNpOr F_[ ;Ou a_[h Eqt v3OvT5Qp3O3sNqpQ QN5u+ ;XXt;RQ N3 v3OspRPNq;N Nqt pOup2pu5;T ;Ou v3TTtvNp2tupPtOQp3O ;Rt O3N v3OsTpvNpOrl 85N 3O Nqt v3ONR;R+ RtXRtQtON N.3 upsstRtON s;vtNQ3sv3PXTt- WO3.TturtP;O;rtPtON XR3vtQQh &5RNqtRP3Rt NqpQ QN5u+ pOupv;NtQ ;O5P8tR3sXpNs;TTQ Q5vq ;QJ Nqt Tt2tT 3s v3PXTt-pN+ ;Ou sR;rPtON;Np3O 3s NqtQtN33TQ P;WtQ Nqt 3XNpP;T5Q;rt upsspv5TNl 3Rf;Ou Nqt XRp2;v+ RpQWQ Nq;N 3RprpO;NtQ sR3P Nqt upsspv5TN+ N3 QtX;R;NtNqt :XtRQ3O;T QXqtRt:sR3P Nqt :v3TTtvNp2t QXqtRt:h

15 1Introduction

Eqt ;u2tON 3s Nqt At8 mh4 q;Q Rt2pN;Tp*tuNqt sptTu 3s _O3.Tturt [;O;rtPtONl8+XR3k 2pupOr ;Ot. XtRQXtvNp2t ;Ou N33TQ 8;Qtu 3O Nqt put; Q5XX3RNpOr Xt3XTtv33XtR;Np3O ;Ou X;RNpvpX;Np3Ol ;vWO3.TturpOr NqtpPX3RN;Ovt 3s Nqt Q3vp;T XR3vtQQ pO Nqt .;+ Nq;N Nqt WO3.TturtpQvRt;Ntu ;Ou P;O;rtu 9[vH4ewh #O NqtAt8 mh4 tR;l WO3.Tturt P;O;rtk PtONl 3R _O3.Tturt [;O;rtPtON mh4 7_[ mh46 3R (ONtRXRpQt mh4 ;Q v3pOtu 8+ a;NRpvW [vHstt9[vH4ewl v;O 8t ;QQ3vp;Ntu .pNq;QtN 3s XR3vtQQtQ;Ou N33TQ ;pPpOr ;N vRt;NpOr ;>v3PP5OpN;Rp;O> tO2pR3OPtONpO.qpvqXt3XTtQ3vp;Tp*tl v3TT;83R;Nt ;Ou Nq5Qv3TTtvNpk 2tT+ X;RNpvpX;NtN3Nqt vRt;Np3O;Ou t-vq;Ort 3s WO3.Tturth[3Rt rtOtR;TT+l Nqt P;O;rtPtON 3s WO3.Tturt q;Q t23T2tu sR3P NqtpOpNp;T put; 3s t-NR;vNpOrlvR+QN;TTp*pOr ;Ou QN3RpOr WO3.Tturt pO u;N;8;QtQpO;s3RP Nq;Nv;O 8t t;QpT+ t-XT3pNtu 8+ 3Rr;Op*;Np3OQl N3 Nqt put; 3s vRt;NpOrNqt v3OupNp3OQ Q5XX3RNpOr ;Ou ;5rPtONpOr Nqt v;X;8pTpNptQ s3R Xt3XTt N3 pONtR;vN .pNqt;vq 3NqtRh E3u;+l Nqt XRpP;R+ R3Tt 3s WO3.Tturt P;O;rtPtON pQ N3 vRt;Nt Nqt v3OupNp3OQ ;Ou vTpP;NtpO.qpvqv3TTtvNp2t pONtTTprtOvt pQ s3QNtRtuh

EqpQ XqtO3PtO3O q;Q Q3Ptq3. 38s5Qv;Ntu Nqt put; Nq;N WO3.Tturt ;TQ3 pOvT5utQ ; XR3s35Ou pOup2pu5;T upPtOQp3Oh CTNpP;NtT+WO3.Tturt pQ P;O;rtu 8+ Nqt pOup2pu5;TQ .q3 vRt;Nt pNl XR3vtQQ pNl X5N pN pO ;XXTpv;Np3Ol85N ;TQ3 t-vq;Ort.pNqNqt 3NqtRQh Eqt P;O;rtPtON 3s WO3.Tturt ;N Nqt pOup2pu5;T Tt2tT RtXRtQtONQ;2tR+ pPX3RN;ONpQQ5t s3R Nqt WO3.Tturt .3RWtRQh #rO3RpOr NqpQ ;QXtvN pQ XR38;8T+;OpPX3RN;ONRt;Q3O3sNqt s;pT5Rt 3s ;spRQN rtOtR;Np3O WO3.Tturt P;O;rtPtON .qpvq.;Q t-XtvNpOr sR3P NqtpR tPXT3+ttQ ;O ;TNR5pQNpv ;NNpN5utl ;Ou pO X;RNpv5T;RNq;N Nqt+ .35Tu P;ut ;2;pT;8Tt N3 Nqt .q3Tt 3Rr;Op*;Np3O Nqt :XRtvp35Q: XtRQ3O;T WO3.Tturt Nq;N Nqt+ q;u ;vv5P5T;Ntu ;Q X;RN 3s NqtpR u;pT+ ;vNp2pNptQ 9(u4jwh

F5Rp35QT+l NqpQ pOup2pu5;T XtRQXtvNp2t 3s WO3.Tturt P;O;rtPtONl .qpvqv;O 8t ;QQ3vp;Ntu N3 atRQ3O;T _O3.Tturt [;O;rtPtON 7a_[6l ;XXt;RQ N3 q;2t Rtvtp2tu RtT;Np2tT+ TpNNTt ;NNtONp3Oh [3Rt QXtvpspv;TT+l ;Rt2pt. 3s Nqt TpNtR;N5Rt pOupv;NtQ Nq;N Nqt NtRP “XtRQ3O;T WO3.Tturt P;O;rtPtON” pQ RtT;Np2tT+ Ot.l t2tO Nq35rqpNQ 3RprpOv;O 8t NR;vtu pO ;.3RWpOr X;XtRsR3P &R;Ou ;Ou $p-3O 9&$LLwh Eqt 5OutRT+pOr put; 3s Q5XX3RNpOr WO3.Tturt P;O;rtPtON ;N Nqt pOup2pu5;TTt2tT pQ NqtRts3Rt RtvtON ;Ou O3N 2tR+ tT;83R;Ntuh HRt2pt. 3s TpNtR;N5Rt ;TQ3 pOupv;NtQ Nq;N Nqt pOsT5tOvt ;Ou pPX;vN 3s a_[ q;Q RtP;pOtu V5pNtTpPpNtuJ st. ;RNpvTtQ pO Nqt TpNtR;N5Rt ;XXt;R N3 v32tR XtRQ3O;T WO3.Tturt P;O;rtPtONl ;vWO3.TturpOr ;TpNNTt pONtRtQN pO Nqt WO3.Tturt P;O;rtPtON v3PP5OpN+hEqt Rt;Q3OQ s3R NqpQ TpPpNtu pONtRtQN P;+ 3RprpO;NtsR3PNqt upsspv5TN+ N3 Q5XX3RNXt3XTtP;O;rpOr NqtpR WO3.Tturt ;N Nqt pOup2pu5;TTt2tTl 3R sR3P Nqtupsspv5TN+ 3s 3Rr;Op*;Np3OQ N3 qtTXpOup2pu5;TQ pO NqpQ X;RNpv5T;R ;QXtvNh HO3NqtR ;OQ.tR .35Tu 8t Nq;N pOup2pu5;T WO3.Tturt P;O;rtPtON pQ ;TRt;u+T;RrtT+ XRtQtON ;Ou Q5XX3RNtu 8+ Nqt O5PtR35Q pOup2pu5;T N33TQ Nq;N tPXT3+ttQ 5Qt 3O ;u;pT+ 8;QpQ Q5vq;Q.3Ru XR3vtQQ3RQl v;TtOu;RQl3R3NqtRatRQ3O;T #Os3RP;Np3O [;O;rtPtON 7a#[6 N33TQ 9bE4dwl 9HX4iwh

16 At8mh4 q;Q P;ut ;2;pT;8Tt ;2;RptN+3sOt. N33TQ Nq;N P;+ NR;OQs3RPNqpQ 2pQp3Oh &pRQNl P;O+ 3s NqtQtX;RNpvpX;N3R+N33TQ v;O ;TQ3 8t v3OQputRtu ;vv3RupOrN3;OpOup2pu5;T XtRQXtvNp2th \tv3Ou 8tv;5QtNqt 5Q;rt 3s NqtQtN33TQ ;N Nqt v3TTtvNp2tTt2tT q;Q Q3Pt pPXTpv;Np3O s3R Nqt pOup2pu5;ThHsNtR;TTl t2tO pO ;Q3vp;T v3ONt-Nl Nqt pOup2pu5;T XtRQXtvNp2t pQ ;TQ3 XRtQtONl;Ou s3R pOQN;Ovtl ;vv3RupOrN3Nqt Q3vp;T t-vq;OrtNqt3R+ 9E_gLw Xt3XTtpONtR;vN .pNq 3NqtRQ 8tv;5QtNqt+ t-XtvN ;XtRQ3O;T 8tOtspNh [3Rt pONtRtQk NpOrT+l .pNqNqt Ot. N33TQl Nqt pOup2pu5;T;Ou v3TTtvNp2t;QXtvNQ 3s WO3.Tturt P;O;rtPtONQq35Tu O3N 8t v3OQputRtu ;Q N3N;TT+ QtX;R;Ntu 85N X;RN 3s ;Q;PtrT38;T XR3vtQQJ pO Nqt At8 mh4l Nqt pOup2pu5;TWO3.Tturt ;Ou Nqt v3TTtvNp2tWO3.Tturt ;Rt Nqt Q;Pt WO3.Tturth&3R pOQN;Ovt pO Nqt v;Qt 3s Q3vp;T 833WP;RWpOrl ;5QtR833WP;RWQ ; RtQ35Rvt pOpNp;TT+s3R qpQ XtRQ3O;T 5Qtl 85N NqpQ 833WP;RWl pOup2pu5;T WO3.TturtlpQNqtO ;TQ3 ;5N3P;Npv;TT+P;ut ;2;pT;8Tt N3 3NqtRQ ;Ou 8tv3Pt ;N Nqt Q;PtNpPt v3TTtvNp2t WO3.Tturth EqpQ v3TTtvNp2tWO3.Tturt .pTTT;NtR t23T2t 2p; Nqt v3PP5OpN+ XR3vtQQ v3Ok NRp85NpOr N3 tORpvq 8;vWNqt pOup2pu5;TWO3.Tturth

Eqt 38YtvNp2t 3s NqpQ X;XtRpQN3pO2tQNpr;Nt Nqt pPXTpv;Np3O 3s Nqt At8 mh4l ;TQ3 RtstRRtu N3 ;Q Nqt \3vp;T At8l 3O a_[h &5RNqtRP3Rtl Nqt ;RNpvTt t-XT3RtQ Nqt RtT;Np3OQqpX 8tN.ttO F3TTtvNp2t_O3.Tturt[;O;rtPtON 7F_[6 ;Ou a_[ ;Ou pO X;RNpv5T;RNqt R3Tt 3s a_[ 3O Nqt \3vp;T At8h EqpQ ;RNpvTt t-XT3RtQ .qtNqtR At8mh4 ;XXTpv;Np3OQ;Rt tsstvNp2t N33TQ s3R P;O;rpOr XtRQ3O;T WO3.Tturt ;Ou upQv5QQtQNqt v5RRtONR3;u8T3vWQ ;Ou 3XX3RN5OpNptQh#O3RutR N3 v3Ou5vN NqtQtpO2tQNpr;Np3OQl ;Rt2pt. 3s Nqt upsstRtON At8 mh4 N33TQ pQ v3PXTtPtONtu .pNq;TpNtR;N5Rt Rt2pt. 3s a_[ RtT;Ntu ;RNpvTtQhEqt Qtv3Ou X;RN 3s Nqt ;RNpvTt vT;QQpsptQAt8mh4 N33TQ ;vv3RupOr N3 NqtpR QXtvpspvR3Tt N3 Q5XX3RN a_[ ;Ou upQv5QQtQNqt RtT;Np3OQqpX a_[ ;Ou F_[h

2Personaland Collective Knowledge Management and Web2.0

EqpQ QtvNp3O ;pPQ N3 XR32put ;8;vWrR35Ou s3R 35R RtQt;Rvq QN5u+h #N spRQN Q5R2t+Q t-pQNpOr .3RW pO Nqt sptTu 3s a_[l F_[ ;Ou At8 mh4 N33TQh

17 2.1 PersonalKnowledge Management(PKM)

_O3.Tturt pQ XtRQ3O;Th#Nv3OQNpN5NtQ ;O pON;Orp8Tt ;QQtN s3R pOup2pu5;TQh_O3.Tturt pQ utspOtu 8+ .q;N .t WO3.l pN v3OQpQNQ 3s NR5NqQl8tTptsQl Y5urPtONQl ‘WO3. q3.’l PtNq3k u3T3rptQl tNvha3T;O+p 9a3Ldw tPXq;Qp*tQNq;N Nqt P3QNpPX3RN;ONN+Xt3sWO3.Tturt pQ XtRQ3O;T ;Ou q;Ru N3 ;RNpv5T;Nth _O3.Tturt pQ ;TQ3 XtRQ3O;T pO Nqt QtOQt Nq;N Xt3XTtq;2t upsstRtON pONtRXRtN;Np3OQ;Ou upsstRtON .;+Q 3s Rt;Q3OpOrh atRQ3O;T WO3.Tturt v;O pOvT5ut WO3.Tturt r;pOtu sR3P PtP3RptQl Rt;upOrQ 833WQl O3NtQl u3v5PtONQl Xq3N3k rR;XqQlpON5pNp3OQl XtRQ3O;Tv3ON;vNQ ;Ou RtT;Np3OQqpXQl .q;N 3Ot q;Q Tt;ROtu sR3P v3TTt;r5tQl;Ou .q;N ;XtRQ3O WO3.Q ;835N t2tR+NqpOr pO Nqt .3RTu 9[;44wh atRQ3O;T WO3.Tturt P;O;rtPtON 7a_[6 v3OQpQNQ pO ;v3TTtvNp3O3sXR3vtQQtQNq;N ;O pOup2pu5;T OttuQ N3 v;RR+35N pO 3RutR N3 r;NqtRl vT;QQps+lQN3RtlQt;Rvql ;Ou RtNRpt2t WO3.k Tturt pO qpQfqtRu;pT+ ;vNp2pNptQ 9%R4dwh atRQ3O;T _[ q;Q 8ttO;XXR3;vqtu sR3P 2tR+ upsstRtON XtRQXtvNp2tQIQ3Pt ;5Nq3RQs3v5Q 3O QXtvp;T;QXtvNQ 3s WO3.Tturt .3RW 3O q3. N3 8tNNtR5NpTp*t ;v3PX5NtR N3 qtTXNqt WO3.Tturt .3RWtR N3 P;O;rtqpQfqtR WO3.Tturt tsstvNp2tT+ 79\[\4mwl 9G\41wl 9);4gwl 9)H]4ewl 9$E41wl 9_341wl 9\vq4gw6l .qpTt 3NqtRQ s3v5Q 3O XR38TtPkQ3T2pOr QWpTTQ 3R X3QQp8pTpN+3s3Rr;Op*pOr put;QtsspvptONT+ 79\[\4mwl 9HT4gwl9H%4ewl 9)%4dwl 9D]\4ew6h )psstRtON sR;Pt.3RWQ ;Ou N33TQ q;2t 8ttOXR3X3QtuN;WpOrpON3 ;vv35ON QXtvpspvvq;R;vNtRpQNpvQ 3s WO3.Tturt .3RW ;uuRtQk QpOrJNqt pOs3RP;Np3O 32tRT3;u XR38TtP 79HE4jwl 9)(4gw6l XtRQ3O;Tp*;Np3Ol v3ONt-N5;Tpk *;Np3O ;Ou v5QN3Pp*;Np3O;QXtvNQ 79$E41wl 9]H4jwl 9];4gw63RsR;Pt.3RWQ 5OutRT+pOr Nqt pPX3RN;Ovt 3s WO3.Tturt Qq;RpOr 79__4ewl 9]HZ4jw6h

2.2 Collective Knowledge Management (CKM)

AqtRt;Q XtRQ3O;T WO3.Tturt pQ v3OvtONR;Ntu 3O Nqt pOup2pu5;Tlv3TTtvNp2tWO3.Tturt RtstRQ N3 WO3.Tturt Nq;N pQ v3PP3ON3;TT PtP8tRQ3s;O3Rr;Op*;Np3O 9%RLewh G;5P;Ru 9G;LLw utspOtQ F_[ ;Q >WO3.Tturt 3s Nqt 5OQX3WtOl 3s Nqt pO2pQp8Tt QNR5vN5Rt>h F3TTtvNp2tWO3.Tturt pQ utspOtu ;Q WO3.Tturt 3s ;O tO2pR3OPtON3stQN;8TpQqtu R5TtQl T;.Q ;OuRtr5T;Np3OQ 9%RLewh

&pr5Rt 1J E+XtQ 3s _O3.Tturt 79G;LLwl Xhee6

18 F_[ RtstRQ N3 ;QtN 3s N33TQ ;Ou XR3vtQQtQNq;N ;Rt 5Qtu N3 P;O;rtNqt v3TTtvNp2t WO3.Tturt 3s ;O 3Rr;Op*;Np3Oh F_[ q;Q RtXRtQtONtu NqtP;pO QN5u+ Q58YtvN 3s WO3.Tturt P;O;rtPtONl ;Ou pN pQ 5Q5;TT+;QQ3vp;Ntu .pNqNR;upNp3O;T WO3.Tturt P;O;rtPtON 9H^41wh #N Qq35Tu 8t O3NtuNq;N NR;upNp3O;T WO3.TturtP;O;rtPtON pOvT5ut 83Nq Nqt P;O;rtPtON 3s N;vpN ;Ou t-XTpvpN WO3.Tturtl85N v3OQputRtu P;pOT+ ;vv3RupOr N3 ;rT38;TXtRQXtvNp2t;Qv;O 8t pTT5QNR;Ntu 8+ Nqt \(F#7\3vp;Tp*;Np3Ol (-NtRO;Tp*;Np3Ol F3P8pO;Np3Ol ;Ou #ONtRO;Tp*;Np3O6 P3utT 3s Z3O;W; ;Ou E;Wt5vqp 9ZELgwh EqpQ P3utT XR3X3QtQWO3.Tturt XR3vtQQtQ ;RNpv5T;NpOr ;Ou v3O2tRNpOr Nqt t-XTpvpN ;Ou N;vpN WO3.Tturt v3O2tRQtT+h

2.3 Web 2.0 and KM 2.0

At8 mh4 9DRt4gw RtstRQ N3 ;Ot. 2pQp3O 3s Nqt At8 8;Qtu 3O Xt3XTtX;RNpvpX;Np3O ;Ou pONtR;vNp3OlpOv3ONR;QN.pNqNqt s3RPtR2pQp3O Nq;N v3OQputRtu Nqt At8 3OT+ ;Q ;q5rt pOs3RP;Np3O QX;vth _[ mh4 3R (ONtRXRpQt mh4 79_;4dwl 9[vH4ewl 9\_4dw6 RtstRQ N3 Nqt ;u;XN;Np3O 3s NqpQ 2pQp3O N3 Nqt P;O;rtPtON 3s WO3.Tturt pO Nqt tONtRXRpQtl ;Ou s3v5QtQ 3O WO3.Tturt t-vq;OrtQ ;Ou v3TT;83R;Np3O ;P3OrQNNqt tPXT3+ttQlNqt WO3.Tturt .3Rk WtRQ 3s Nqt 3Rr;Op*;Np3Oh EqpQ P3utTNRptQ N3 q;ROtQQ Nqt v3TTtvNp2tpONtTTprtOvt .qpTt ;N Nqt Q;PtNpPt ;vvtTtR;Nt Nqt vpRv5T;Np3O 3s WO3.Tturt ;P3OrQN Xt3XTth [3Rt QXtvpspk v;TT+l pNQ RtTptQ3ONqt put; 3s qtTXpOr v3PX;OptQ N3 8tNNtRt-XT3pN Nqt pON;Orp8Tt WO3.k Tturt;QQtNQ Nq;N pQ XRtQtON pO Nqt Xt3XTtqt;u8+tPX3.tRpOr Nqt pOup2pu5;TQl8+P;WpOr t;vq 3s NqtP ;s5TT v3ONRp85N3R;Ou 8+ s;vpTpN;NpOr vpRv5T;Np3O 3s NqtpR WO3.Tturth aR;vNpv;TT+l At8 mh4 ;Ou (ONtRXRpQt mh4l q;Q pONR3u5vtu ;2;RptN+3sN33TQ N3 Q5XX3RN NqpQ 2pQp3O Q5vq ;Q 8T3rQl .pWpQl ;Ou Q3vp;T OtN.3RWQl ;Q .tTT;Q;O5P8tR3sPtvq;OpQPQ PtNq3uQ ;Ou XRpOvpXTtQ N3 8t 5Qtu s3R Q5XX3RNpOr ;Ou tOv35R;rpOrv3TT;83R;Np3OhEqtRt q;Q 8ttO;R;Xpu;u3XNp3O3sAt8 mh4 N33TQ ;Ou ;XXTpv;Np3OQ;vR3QQ 3Rr;Op*;Np3O 85N .pNq upsstRtON ;u3XNp3O X;NNtROQ utXtOupOr 3s Nqt rt3rR;Xqpv;T T3v;Np3O3Rv5TN5R;T s;vN3RQ 9[v_4dwl s3R pOQN;Ovt Q3Pt v35ONRptQQ5vq ;Q FqpO; 8tpOrP3RtpONtRtQNtu pO Nqt 5Qt 3s Q3vp;T OtN.3RWQ Nq;O 3NqtRQh

3Web 2.0 Tools for Supporting PKMand CKM

#O Nqt s3TT3.pOr Q5R2t+l .t .pTTpO2tQNpr;Nt q3. At8 mh4 ;XXTpv;Np3OQQ5vq ;Q 8T3rQl Q3k vp;T OtN.3RWpOr N33TQ 3R .pWpQv;O v3ONRp85Nt N3 Nqt P;O;rtPtON 3s XtRQ3O;T ;Ou v3TTtvNp2tWO3.Tturth

3.1 ASurvey of Web 2.0 Tools

Eqt RtQ5TN3sNqpQ Q5R2t+ v3OQpQNQ pO ;O pO2tON3R+ 3s At8 mh4 N33TQl ;Ou NqtpR ;XXTpv;8pTpN+ s3R Nqt P;O;rtPtON 3s XtRQ3O;T ;Ou v3TTtvNp2tWO3.Tturth

19 aR;vNpv;TT+Nqt PtNq3u3T3r+;u3XNtu s3Rv3Ou5vNpOr NqpQ Q5R2t+ q;Q QN;RNtu8+;TpNtR;N5Rt Rt2pt. 3s ;RNpvTtQ v3OQputRpOr Nqt 5Qt 3s At8 mh4 N33TQ pO ;O tONtRXRpQt v3ONt-Nl thrh 9DBG4ewh EqpQ spRQN Xq;Qt.;Q 5Qts5T N3 putONps+ Nqt upsstRtON v;Ntr3RptQ 3s At8 mh4 N33TQ Q5vq ;Q 3s .pWpQ3R8T3rQ 9)t4dw Nq;N ;Rt 5Qtu N3 Q5XX3RN Nqt WO3.Tturt P;O;rtPtON XR3vtQQ pO tONtRXRpQtQhAtq;2t Qt;Rvqtu;Ou vT;QQpsptu At8 mh4 N33TQ Nq;N v;O 8t 5Qtu s3R a_[ ;Ou F_[h#O;XRt2p35Q ;RNpvTt .t pONtR2pt.tu tPXT3+ttQ sR3P upsstRtON v3PX;OptQ RtT;Ntu N3 Nqt 5Qt 3s At8 mh4 N33TQ pO 3Rr;Op*;Np3OQ ;Ou upQv5QQtu Ot. s3RPQ 3s pONtR;vNp3O ;Ou WO3.Tturt Qq;RpOr 9_]\4MwhEqt pO2tQNpr;Ntu v;Qt QN5uptQ .tRt P;pOT+ #E 85QpOtQQtQ;Ou P;ut 5Q uR;. Nqt XRtTpPpO;R+v3OvT5Qp3O Nq;N Nqt t;RT+ ;u3XNtRQ3s_[mh4 ;Rt P;pOT+ #E 85QpOtQQtQh

At q;2t ;O;T+*tu NqtQt N33TQ ;vv3RupOr N3 NqRtt P;pO upPtOQp3OQJNqt Q5XX3RN3s WO3.Tturt .3RWtRQ N;QWQl NqtpR P;pO s5OvNp3O;TpNptQ ;Ou NqtpR R3Tt pO Nqt v3ONt-N 3s a_[ ;Ou F_[hE;8Tt 1XR32putQ;vT;QQpspv;Np3O3sNqt upsstRtONv;Ntr3RptQ 3s At8 mh4 N33TQl ;Ou utspOtQ NqtpRR3Tt pO Nqt v3ONt-N 3s a_[ ;Ou pO Nqt v3ONt-N 3s F_[h

Web 2.0 Examples Role in PKM Role in CKM tools context context category AggregJ ZtN2p8tQf%pOrtRJ HrrRtr;Ntupsstk HrrRtr;Nt upsstk atRQ3O;Tp*tu ...hOtN2p8tQhv3Pla;rt&T;WtJ RtON Q35RvtQ 3s RtON Q35RvtQ 3s .t8 X;rtQ ...hX;rtsT;WtQhv3Plp%33rTtJ pOs3RP;Np3O pOs3RP;Np3O ;Ou pOs3RP;k ...hr33rTthv3Pfprl[+=;q33@J RtTt2;ON N3 Nqt RtTt2;ON N3 Nqt Np3O P+h+;q33hv3P pOup2pu5;T pO ; v3PP5OpN+ pO ; ;rrRtr;N3RQ QpOrTt XT;vt 7Rtk QpOrTtXT;vt u5vNp3O 3s v3Pk XTt-pN+6 PSearchJ \.pvWpJ...hQ.pvWphv3P `5pvWXtRQ3O;Tpk \q;Rt WO3.Ttuk atRQ3O;Tp*tu *tu Qt;Rvq s3R rt .pNqNqt Qt;Rvq N3XpvQ 3NqtRQ X3RN;TQ VComJ E;OrTtRJ...hN;OrTtRhv3P FRt;Nt 3.O upQk \q;Rt WO3.Ttuk atRQ3O;Tp*tu v5QQp3O s3R5PQl rt ;Ou 3XpOp3OQ Tp2t v3TTtvN 3XpOp3OQl .pNqNqt 3NqtRQ upQv5QQp3O ;QQtQQ pOs3RP;k s3R5PQ ;Ou Np3O 3O RtX5N;k v3PP5OpNptQ Np3O VWorldJ \tv3Ou ^pstJQtv3OuTpsthv3Pl DRr;Op*t;Ou HTT3. N3 t-XtRpk BpRN5;T B;QNX;RWJ...h2;QNX;RWhv3Pl 2pQ5;Tp*t v3Ok PtON ;Ou pONtRk .3RTuQ `.;VJ ...hV.;Vhv3P NtON ;vN .pNqNqt 3NqtRQ BlogJ GT3rrtRJ...h8T3rrtRhv3Pl ]tv3Ru XtRQ3O;T F3TTtvNl Qq;Rt At8T3rQ A3RuaRtQQJ .3RuXRtQQhv3Pl t-XtRptOvtQ ;Ou t-XtRptOvtQ ;Ou E+XtX;uJ...hN+XtX;uhv3P XR32put Nqt 3Xk 3XpOp3OQ X3RN5OpN+ N3 s3Rk P5T;Nt 3XpOp3OQ

20 Wiki [tup;ApWpJApWpXtup;hv3Pl HTT3.Q N3 Rtk FRt;Nt ;Ou P;k [3pO[3pO ApWpJP3pOP3hpOl v3Ru;Ou 3Rr;Opk O;rt pOs3RP;k GR;pOWttXtRJ *t XtRQ3O;T Np3O v3TT;83R;Npk ...h8R;pOWttXtRhv3Pl WO3.Tturt 2tT+ 5QpOr Eq35rqN&;RPtR J q+XtRTpOWtu ...hNq35rqNs;RPtRhv3P QNR5vN5RtQ

Tag: )tThpvp3h5QJ utTpvp35Qhv3Pl F3TTtvNl F3TTtvN ;Ou \3vp;T 833Wk \N5P8TtCX3OJ ;OO3N;Nt Qq;Rt RtstRtOvtQ P;RWpOr ...hQN5P8Tt5X3Ohv3Pl )prrJ RtstRtOvtQ .pNq3NqtRQ uprrhv3Pl FpNtC^pWtJ ...hvpNt5TpWth3Rr OSNJ ^pOWtu#OJ ...hTpOWtupOhv3Pl [;O;rt XtRQ3k F3OOtvN.pNq DOTpOt &;vt833WJ ...hs;vt833Whv3Pl O;T ;Ou XR3stQk 3NqtRXt3XTtl Q3vp;T ZpOrJ ...hOpOrhv3PlsR;Ptk Qp3O;TRtT;Np3Ok ut2tT3X ;O 3Ok OtN.3RWpOr .3RWQQ5vq ;Q DXtO\3vp;T QqpXQl P;Wt Nqt TpOt putONpN+ v3uthr33rTthv3Pf;XpQf3XtOQ3vp;Tf 3.O t-XtRNpQt &RptOu F3OOtvNJ P3Rt2pQp8Tt ...hr33rTthv3PfsRptOuv3OOtvN 7Q3vp;T v;XpN;T6 E;8Tt 1J D2tR2pt.3sAt8 mh4 a_[ N33TrR35XQ

#O 35R Q5R2t+ .t s35Ou ;2;RptN+3sN33TQ Q5XX3RNpOr ;T;Rrtup2tRQpN+ 3s XR3vtQQtQ;Ou s5OvNp3O;TpNptQl;QutQvRp8tu pO E;8Tt 1h Z3 QpOrTtQ+QNtPv32tRQ Nqt .q3Tt R;Ort 3s XtRQ3O;T WO3.Tturt P;O;rtPtON XR3vtQQtQh EqpQ QpN5;Np3O X;RNp;TT+ 3RprpO;NtQ sR3P Nqt 2tR+ P3u5T;R ;Ou upQNRp85Ntu O;N5Rt 3s NqtQtN33TQhAqtRt;Q Nqt R3Tt 3s Nqt N33TQ pO a_[ pQ 3O Nqt pOup2pu5;T Tt2tT 85N ;TQ3 Q5XX3RNQ v3TT;83R;Np3OlNqtpR R3Tt pO F_[ pQ v3OvtONR;Ntu 3O v3TT;83R;Np3Oh

3.2 Analyzing the Effectiveness of Web 2.0 Tools for SupportingPKM

E3 s5RNqtRv3PXTtNt 35R ;O;T+QpQl.tT33Wtu ;N .qpvqt-NtOuNqt upsstRtON At8 mh4 N33TQ Q5XX3RN Nqt upsstRtON a_[ XR3vtQQtQh At .35Tu TpWt N3 utP3OQNR;NtqtRt Nq;N At8 mh4 >Q3vp;T>N33TQ v;O ;TQ3 8t ;XXTptu ;N Nqt pOup2pu5;T Tt2tTl ;Ou ;vN5;TT+ RtXRtQtON 2tR+ 2;T5;8Tta_[ N33TQh

At 8;Qtu 35R ;O;T+QpQ 3O Nqt .3RW 3s H2tR+ tNh;Th 9HG41w .qpvqutspOt Qt2tOXtRQ3O;T _[ P;pO vq;R;vNtRpQNpvQ 3R QWpTTQ Nq;N ;Rt5Qts5T s3R Q5XX3RNpOr XtRQ3O;T WO3.Tturt P;O;rtPtONJ

 ]tNRpt2pOr7Qt;RvqpOr ;Ou putONpspv;Np3O6  (2;T5;NpOr 7;QQtQQpOr Nqt V5;TpN+;Ou RtTt2;Ovt 3s pOs3RP;Np3O6  DRr;Op*pOr pOs3RP;Np3O  F3TT;83R;NpOr ;R35OupOs3RP;Np3O  HO;T+*pOr ;Ou P;WpOrQtOQt 3s pOs3RP;Np3O  aRtQtONpOr pOs3RP;Np3O

21  \tv5RpOr pOs3RP;Np3O 73R PtP3Rp*pOr pOs3RP;Np3O6

[3Rt QXtvpspv;TT+l s3R t;vq v;Ntr3R+ 3s N33T.tNRptu N3 utNtRPpOt .qpvq XtRQ3O;T_[ vq;R;vNtRpQNpvQ .tRt Q5XX3RNtuh E;8Tt mQ5PP;Rp*tQNqt vq;R;vNtRpQNpvQ 3s t;vq v;Ntr3R+ 3s N33TQ N;WpOrpON3 ;vv35ON Nqt H2tR+ vT;QQpspv;Np3O9HG41wh 7? pOupv;NtQ ;vq;R;vNtRpQNpv pQ QNR3OrT+ Q5XX3RNtul .qtRt;Q ;O -pOupv;NtQ ;vq;R;vNtRpQNpv pQ Q5XX3RNtu ;N ;vtRN;pO Tt2tT 3OT+6h

PKM Skills Aggreg PSearch VCom vWorld Blogs Wiki Tag OSN

]tNRpt2pOr - ? - - - - pOs3RP;Np3O

(2;T5;NpOr ? - ? pOs3RP;Np3O

DRr;Op*pOr ? - ? ? ? pOs3RP;Np3O

F3TT;83R;k - - - - - ? ? - NpOr

HO;T+*pOr ? - ? - pOs3RP;Np3O aRtQtONpOr ? - ? pOs3RP;Np3O

\tv5RpOr ? ? - - pOs3RP;Np3O

E;8Tt 1J \5XX3RN 3s Nqt a_[ QWpTTQ 8+ Nqt upsstRtON v;Ntr3RptQ 3s At8 mh4 N33TQ HTT NqtQtv;Ntr3RptQ 3s At8 mh4 N33TQ Q5XX3RN v3TT;83R;Np3O ;Ou ;vtRN;pO Q3vp;T pONtR;vNp3O .pNqXt3XTt.qpvqt-XT;pOQ .q+ At8 mh4 pQ 3sNtO RtstRRtuN3;QNqt \3vp;T .t8h ApNq At8 mh4 XtRQ3O;T WO3.Tturt pQ RtNRpt2tul;O;T+*tu ;Ou XRtQtONtu O3N 3OT+ s3R ;O pOup2pu5;T 5Qtl 85N pN pQ ;TQ3 ;2;pT;8Tt ;N Nqt v3PP5OpN+ Tt2tT s3R ;v3TTtvNp2t5Qth Eqt pOup2pu5;T WO3.Tturt v;O 8t ut2tT3Xtus5RNqtR8+3NqtRX;RNpvpX;ONQl pOvRt;QtQ Nqt WO3.Tturt 3s Nqt pOup2pu5;T ;Ou Nqt v3PP5OpN+ ;Ou qtTXQ N3 85pTu;v3TTtvNp2t pONtTTprtOvth

$3.t2tR pN Qq35Tu 8t O3NtuNq;N Nqt O;N5Rt 3s Nqt pOs3RP;Np3O Nq;N pQ RtstRRtu N3 pQ 2tR+ upsstRtON sR3P ;v;Ntr3R+ 3s N33TN3;O3NqtRh&3R pOQN;OvtlpONqt v;Qt 3s Nqt ApWp 3R E;r NqpQ pOs3RP;Np3O v3OQpQNQ P;pOT+ pO u3v5PtONQl .qtRt;Q pO Nqt v;Qt 3s D\Zl NqpQ pOs3RP;Np3O RtstRQ N3 Xt3XTtXR3spTtpOs3RP;Np3Oh

22 4Discussion and Outlook

At8 mh4 XT;+Q ;P5TNps;vtNtu R3Tt s3R v3PP5Opv;Np3Ol v3TT;83R;Np3Ol WO3.Tturt Qq;RpOr ;Ou WO3.Tturt P;O;rtPtONhAt8 mh4 N33TQ tO;8Tt ;Ot. P3utT 3s XtRQ3O;T _[ Nq;N q;ROtQQtQ Nqt v3TTtvNp2tWO3.Tturt NqR35rq s3RP;T;Ou pOs3RP;T v3PP5Opv;Np3Ol v3TT;k 83R;Np3O ;Ou Q3vp;T OtN.3RWpOr N33TQh EqpQ Ot. XtRQ3O;T _[ s;vpTpN;NtQ pONtR;vNp3Ol v3TT;83R;Np3O ;Ou WO3.Tturt t-vq;OrtQ 3O Nqt At8 ;Ou pO 3Rr;Op*;Np3OQh _[ 1h4 .;Q tQQtONp;TT+ RtT+pOr 3O Nqt put; 3s t-NR;vNpOr WO3.Tturt sR3P Nqt Xt3XTt;Ou P;WpOrpN ;2;pT;8Tt N3 Nqt rR35Xl ;Ou Nq5QupQRtr;RupOr Nqt 2tR+ XtRQ3O;T O;N5Rt 3s WO3.TturthEqt ;5Nq3RQ3sNqpQ ;RNpvTt 8tTpt2t Nq;NNqt P;O;rtPtON 3s WO3.Tturt RtT+pOr At8 mh4 XRpOvpXTtQ s3TT3.Q ;R;upv;TT+upsstRtON X;NqNq;N v;O X3NtONp;TT+ P;Wt Nqt P;O;rtPtON 3s WO3.Tturt P3RttsspvptON 83Nq ;N pOup2pu5;T ;Ou ;N v3TTtvNp2tTt2tTh #O X;RNpv5T;Rl NqpQ X;XtRq;Q Qq3.O Nq;N Nqt P;O;rtPtON3sXtRQ3O;T WO3.Tturt ;Ou a_[ ;Rt O3N pO 3XX3QpNp3O 3R prO3R;ON N3 Nqt Q3vp;T 2pQp3O 8R35rqN 8+ Nqt At8 mh4l 85N pQ ;8Tt N3 Q5XX3RN pN P3RttsstvNp2tT+h

EqpQ X;XtRq;Q 32tR2pt.tu Nqt 5Qt 3s At8mh4 N33TQ s3R Nqt P;O;rtPtON 3s XtRQ3O;T WO3.Tturt ;Ou v3TTtvNp2tWO3.Tturth#OX;RNpv5T;R.tq;2t ;O;T+*tu upsstRtON At8 mh4 ;XXTpv;Np3OQQ5vq ;Q GT3rQl \3vp;T ZtN.3RWQl ApWpQltNvhl ;Ou .t q;2t vT;QQpsptu NqtP 8;Qtu 3O Nqt QXtvpspvR3Tt 3R s5OvNp3O;TpN+ Nqt+ Q5XX3RN ;Q RtXRtQtONtu pO E;8Tt1 ;Ou E;8Tt mh At8 mh4 N33TQ tOv3PX;QQ;2;RptN+3sN33TQl t;vq v;Ntr3R+ v3ONRp85NtQ N3 Q5XX3RN ;X;RNpv5T;R;QXtvN 3s Nqt XtRQ3O;T WO3.Tturt P;O;rtPtONh At8 mh4 N33TQ v32tR .tTTNqt upsstRtON s;vtNQ3sa_[ ;Q pOupv;Ntu pO N;8Tt mh H2;RptN+3sN33TQ P;+ 8t 5Qtu N3 8tNNtRP;O;rt Nqt Q3vp;T v;XpN;T 7.pNqQ3vp;T OtN.3RWpOr Q+QNtPQ Q5vq ;Q ^pOWtu#O6lN3 qtTXN3v3PP5Opv;Nt P3RttsstvNp2tT+ .pNq3NqtRQ 7.pNqXtRQ3O;T 8T3rQl pOQN;ON PtQQ;rpOrQ+QNtPQ6 ;Ouf3R N3 q;ROtQQ v3TTtvNp2tpONtTTprtOvt 7.pNq Q+QNtPQ Q5vq ;Q ApWpQ ;Ou Q3vp;T 833WP;RWpOr6h HTNq35rq At8 mh4 N33TQ q;2t 8ttOutQprOtu;Ou ;Rt rtOtR;TT+ ;QQ3vp;Ntu N3 Q5XX3RN F_[lNqt+ v;O ;TQ3 8t tsstvNp2t ;N Q5XX3RNpOr a_[l ;Ou pO X;RNpv5T;R Nqt P;O;rtPtON 3s WO3.Tturt ;N Nqt pOup2pu5;TTt2tTh

$3.t2tR;vN5;T At8mh4 N33TQq;2t ;O5P8tR 3s TpPpN;Np3OQ;Q.tTTh At8 mh4 N33TQ tOk v3PX;QQ ;QtN 3s QtX;R;Ntu N33TQ utupv;NtuN3Q5XX3RN QXtvpspvs5OvNp3O;TpNptQl 85N Q3 s;R O3 Q+QNtPpONtrR;NtQ ;TT putONpsptu s5OvNp3O;TpNptQh at3XTtq;2t N3 Tt;RO ;QXtvp;TQ+ON;- 7thrh pO ApWpQ63Rq;2t N3 pOQN;TT Nqt Q3sN.;Rt s3R NqtpR 5Q;rt pO Nqt pONR;OtNh HOu 32tR NpPtl pN pQ QNpTT ;XR38TtPN33Rr;Op*t ;Ou vT;QQps+ +35R 3.O v3ONRp85Np3OQ pO 3RutR N3 8t ;8Tt N3 RtNRpt2t NqtP s3R T;NtR 5Qth[;O+ 5QtRQ .q3 v3TTtvN ;Ou P;O;rtNqtpR WO3.Tturt .pNqAt8 mh4 N33TQ ;Rt 5O;.;Rt Nq;N NqpQ WO3.Tturt v35Tu 8t s5RNqtR5Qtu8+3NqtRQh DNqtRXt3XTt3RtONtRXRpQtQ v35Tu Rt5Qt pNl PpQ5Qt pN 3R v3TTtvN pOs3RP;Np3O ;835N Nqt 5QtRQh &5N5Rt _[ Q+QNtP.pTTQ5XX3RN 83NqNqt pOup2pu5;T ;Ou Nqt v3TTtvNp2t upPtOQp3OQl ;Ou .pTT8t;8Tt N3 ;RNpv5T;Nt q;RP3Op35QT+NqtQt N.3 upPtOQp3OQh

CQpOr N33TQ 83Nq s3R ;O pOup2pu5;T ;Ou ;v3TTtvNp2t 5Q;rt ;TQ3 R;pQtQ ;O pPX3RN;ONpQQ5t RtT;Ntu N3 XRp2;v+hEqtRt pQ Nqt RpQW Nq;N WO3.Tturt Nq;N pQ 3RprpO;TT+ P;O;rtu ;N Nqt pOup2pu5;T Tt2tT 3OT+l :Tt;WQ: sR3P Nqt XtRQ3O;T QXqtRt N3 Nqt v3TTtvNp2tQXqtRt ;OupQ P;ut ;2;pT;8Tt N3 Nqt 3NqtRQ .pNq35N Nqt WO3.Tturt 3s Nqt XtRQ3Oh &3R pOQN;Ovt Qt2tR;T QN3RptQ q;2t 8ttORtX3RNtu 3s Xt3XTtq;2pOr T3QNNqtpR Y38 ;sNtR .RpNpOr 3O NqtpR:XtRQ3O;T

23 8T3r: 9Bpo4gwl Nqt XtRQ3OQ q;2pOr s3Rr3NNtO Nq;N Nqt pOs3RP;Np3O X58TpQqtu pO NqtpR 8T3r .;Q ;TQ3 ;vvtQQp8Tt 8+ NqtpR tPXT3+tRhDOTpOt Q3vp;T OtN.3RW ;Rt ;TQ3 XT;vtQRtXRtQtONpOr vtRN;pO RpQWQ N3 NqtpR 5QtRl RtT;Ntu N3 Nqt upQvT3Q5Rt 3s XtRQ3O;T pOs3RP;Np3O 9%R4gwhEqtRt pQ O3 r5;R;ON+ Nq;N NqtpR 5NpTp*;Np3O pO ;O pONR;OtN pO Nqt v3ONt-N 3s ;v3RX3R;Nt WO3.Tturt P;O;rtPtON pOsR;QNR5vN5RtP;+ N3N;TT+ ;TTt2p;Nt NqpQ XR38TtPh &3R pOQN;Ovtl NqtRt;Rt Q3Pt RpQWQ Nq;N pOs3RP;Np3O ;835N Q3vp;T RtT;Np3OQqpXQNq;N ;XtRQ3O Rtv3RuQ s3R pNQ XtRQ3O;T 5Q;rtl 3R s3R ;vT3Qtu v+vTt 3s sRptOuQl pQ P;ut ;2;pT;8Tt N3 ;T;RrtR ;5uptOvt 8tv;5Qt3sQ3Pt pO;utV5;Nt v3Ospr5R;Np3O ;Ouf3R ;.R3Or sttTpOr 3s Qtv5RpN+h

References

9HG41w H2tR+l \hl GR33WQI ]hl GR3.Olbhl )3RQt+l ahl D: F3OOtRl [hJ atRQ3O;T _O3.Tturt [;O;rtPtONJ&R;Pt.3RW s3R #ONtrR;Np3O ;Ou a;RNOtRQqpXQh #O H\FC(lm441h 9H%4ew HXQq2;TW;l)hl %R5OuQXtOWpQl bhJ atRQ3O;T WO3.Tturt P;O;rtPtON ;Ou pONtTTprtON ;rtONXtRQXtvNp2th #O 7tuh ZpTQQ3Ol Hh %h tN ;Th6 aR3vttupOrQ 3s Nqt 1iNqpONtRO;Np3O;T v3OstRtOvt 3O pOs3RP;Np3O Q+QNtPQ ut2tT3XPtONl_;RTQN;ul \.tutOl m44eh 9H^41w HT;2pl [hl ^tpuOtRl )h(hJ ]t2pt.J_O3.Tturt [;O;rtPtON ;Ou _O3.Tturt [;O;rtk PtON \+QNtPQl [#\ `5;RNtRT+ 7mgJ16l [;Rvqm441l XXh 14dk1jeh 9HT4gw HTXtRNl \h ]hhJF3PXRtqtOQp2t P;XXpOr 3s WO3.Tturt ;Ou pOs3RP;Np3O RtQ35RvtQJ Eqt v;Qt3sAt8QNtRh #O _O3.Tturt ;Ou #Os3RP;Np3O BpQ5;Tp*;Np3OJ \t;RvqpOr s3R \+OtRrptQl m44gh 9HX4iw HXQq2;TW;l)hJ atRQ3O;T _O3.Tturt [;O;rtPtONh#OJ ]([(Z=#l );O 7tuQh6J aR3vh 3s 11Nq (5R3Xt;OF3OstRtOvt 3O #Os3RP;Np3O EtvqO3T3r+ (2;T5;Np3Ol]3+;T ZtNqtRT;OuQ Hv;utP+3sHRNQ ;Ou \vptOvtQl HPQNtRu;Pl NqtZtNqtRT;OuQl 11 k1mZ32hl (5R3Xt;O F3Osh 3O #Os3RP;Np3O EtvqO3T3r+ (2;T5;Np3Oh]t;upOrl C_ JHv;utPpv F3OstRtOvtQ ^pPpNtul m44il XXh 1dkmmh 9HE4jw HT2;R;u3l Fhl Ett2;Olbhl HvWtRP;OOl [h\hl _;RrtRl )hJ \5R2p2pOr Nqt #Os3RP;Np3O (-XT3Qp3OJ$3. at3XTt &pOu EqtpR (TtvNR3Opv #Os3RP;Np3Oltuh H# [tP3j44mk44eJ [;QQ;vq5QtNNQ #OQNpN5Nt3sEtvqO3T3r+lHRNpspvp;T#ONtTTprtOvt^;8hlm44jh 9G;LLwG;5P;Rul ahJ E;vpN _O3.Tturt pO DRr;Op*;Np3OQl \;rtl^3Ou3Ol 1LLLh 9G\41wGtOQ3Ol\hl \N;OupOrl FhJ (sstvNp2t WO3.TturtP;O;rtPtONJ _O3.TturtlNqpOWpOr;Ou Nqt XtRQ3O;Tv3RX3R;NtWO3.Tturt Ot-5Q XR38TtPl#Os3RP;Np3O \+QNtPQ &R3ONptRQ j7m6JmmdkmjMl m441h 9)(4gw );TQr;;Rul ahl (RpWQQ3Ol(hl _3ts3tu $;OQtOl ^hJ ]tNqpOWpOr pOs3RP;Np3O q;OuTpOrJ utQprOpOr s3R pOs3RP;Np3O 3ssT3;ul #O aR3vttupOrQ 3s Nqt iNqutvtOOp;T v3OstRtOvt 3O FRpNpv;T v3PX5NpOrJ 8tN.ttOQtOQt ;Ou QtOQp8pTpN+h H;Rq5Ql )tOP;RWJ HF[l m44gh 9);4gw );2tOX3RNl Eqh $hJ EqpOWpOrs3R ;^p2pOrJ $3. N3 %tN GtNNtR atRs3RP;Ovt ;Ou ]tQ5TNQ sR3P _O3.Tturt A3RWtRQl G3QN3OJ $;R2;Ru G5QpOtQQ \vq33T a58TpQqpOrl m44gh 9)H]4ew);2ptQl \hl HTTtOl \hl ];Xq;tTQ3Olbhl [tOrl (hl (OrTtP;Ol bh _pOrl ]hl ^t.pQlFhJ a3Xv3ROJ Nqt XtRQ3O;TWO3.Tturt 8;Qth#OaR3vttupOrQ 3s Nqt eNqv3OstRtOvt3O)tQprk OpOr #ONtR;vNp2t Q+QNtPQl COp2tRQpN+ a;RWl aHl C\HJ HF[l m44eh 9)t4dw )t;RQN+Otl GhAhJ Huu Huutu GT3rQl [;Qq5XQl

24 9%RLew %R;ONl ]h[hJ E3.;Ru ;WO3.Tturtk8;Qtu Nqt3R+ 3s Nqt spRPh\NR;Ntrpv [;O;rtPtON b35RO;T 1dl XXh 14Lk1mml 1LLeh 9%R4gw %R3QQl ]hl HvV5pQNpl Hhl $tpO*l $h bhJ #Os3RP;Np3O Rt2tT;Np3O ;Ou XRp2;v+ pO 3OTpOt Q3vp;T OtN.3RWQh #O aR3vttupOrQ 3s Nqt m44g HF[ A3RWQq3X3OaRp2;v+pONqt (TtvNR3Opv \3vptN+7HTt-;OuRp;l BHl C\Hl Z32tP8tR4dk4dl m44g6h Aa(\ :4gh HF[l Zt. =3RWl Z=l d1kM4 9%R4dw %R5OuQXtOWpQlbhJ HrtON 8;Qtu;XXR3;vq s3R 3Rr;Op*;Np3O ;Ou XtRQ3O;TWO3.Tturt P3utTTpOrJ WO3.Tturt P;O;rtPtON XtRQXtvNp2tl b35RO;T 3s #ONtTTprtON [;O5s;vN5RpOr 1M7i6Jig1kigdl m44dh 9$E41w $pvWQl )hl E3vqNtRP;OOl _hJ atRQ3O;T uprpN;T Tp8R;RptQ ;Ou WO3.Tturt P;O;rtPtONl b35RO;T 3s COp2tRQ;TF3PX5NtR \vptOvtd7d6Jgg4kgegl m441h 9bE4dw b3OtQlAhl Ett2;OlbhJ atRQ3O;T pOs3RP;Np3O P;O;rtPtONh \t;NNTtlA;QqhJ COp2h 3s A;Qk qpOrN3O aRtQQl m44dh 9_;4dw _;Wp*;.;l =hJ #Okq35Qt 5Qt 3s At8 mh4J (ONtRXRpQt mh4l Ztv EtvqOpv;T b35RO;T m7m6Jiek iLl m44dh 9__4ew _pPl)h$hl _pPl\hbhJ &R;Pt.3RW s3R v3TT;83R;Np2t WO3.Tturt Qq;RpOr ;Ou Rtv3PPtOk u;Np3O 8;Qtu3ON;-3O3Ppv X;RNp;TRtX5N;Np3OQl #O _O3.Tturt \vptOvtl (OrpOttRpOr ;Ou [;O;rtPtONlm44eh 9_341w _3qTl &hJ #Oup2pu5;T WO3.Tturt P;O;rtPtONk\NR;NtrptQ s3R XtRQ3O;Tq;OuTpOr3spOs3Rk P;Np3O ;Ou WO3.Tturt ;N Nqt .3RWXT;vtl aQ+vqp;NRpQvqt aR;-pQ mM7m6Jdikdil m441h 9_]\4Mw _pRvqOtRl _hl ];*PtRpN;l ^hl \5u*pO;l&hJ Zt. &3RPQ 3s #ONtR;vNp3O ;Ou _O3.Tturt \q;RpOr 3O At8 mh4l #OJ ^+NR;Ql [hl );Pp;Opl (hl DRu3Ot* )t a;8T3Qlah7tuQh6J At8mh4J EqtG5QpOtQQ [3utTl \XRpOrtR \vptOvt;Ou G5QpOtQQ [tup;lC\Hl m44Mh 9[;44w [;RNpOl bhJ _O3.Tturt [;O;rtPtONhEqt G;QpQ3sF3RX3R;Nt;Ou #OQNpN5Np3O;T_O3.Ttuk rt [;O;rtPtONh tuh [;RNpOl bhl ARprqNl _hl m444h 9[vH4ew [vHsttl Hh ahJ (ONtRXRpQt mh4J Eqt );.O 3s (PtRrtON F3TT;83R;Np3Ol \T3;O [;O;rtPtON ]t2pt.l\XRpOr m44el B3Th idl Z3h jl XXh m1kmMl m44eh 9[v_4dw [v_pOQt+l $3. G5QpOtQQtQ ;Rt CQpOr At8 mh4l [v_pOQt+ `5;RNtRT+l [;Rvql m44dh 9ZELgw Z3O;W;l #hl E;Wt5vqpl $hJ Eqt WO3.TturtkvRt;NpOr v3PX;O+hZt. =3RWl D-s3RuJ D-s3Ru COp2tRQpN+ aRtQQl 1LLgh 9DBG4ewDRtO (l B0TWtT [l GRtQTpOb%l )tvWtR \h \tP;ONpv ApWpQ s3R atRQ3O;T _O3.Tturt [;O;rtPtONh1dNq#ONtRO;Np3O;TF3OstRtOvtl )(?H m44el _R;W3.l a3T;Oul \tXNtP8tR ikMl aR3vttupOrQh GtRTpO $tputT8tRrJ\XRpOrtRI XXh g4Lkg1Ml m44eh 9DRt4gw D:]tpTT+l EhJ Aq;N #Q At8 mh4 )tQprO a;NNtROQ ;Ou G5QpOtQQ [3utTQ s3R Nqt Zt-N %tOtR;Np3O 3s \3sN.;Rt 8+ EpP D:]tpTT+l 4Lfj4fm44glm44gl ...h3RtpTT+OtNhv3PfX58f;f3RtpTT+fNpPfOt.Qfm44gf4Lfj4f.q;NkpQk.t8km4hqNPT 9D]\4ew D:]p;pOl \hl \X+OQl ahJ (Oq;OvpOrNqt 85QpOtQQ ;O;T+QpQ s5OvNp3O .pNqQtP;ONpvQl #OJ DO Nqt [32t N3 [t;OpOrs5T #ONtROtN\+QNtPQ m44eJ F33XpQl )3;l %;u;l;Ou Du8;Ql aN 1l aR3vttupOrQl m44eh 9a3Ldw a3T;O+pl [h tuhJatRQ3O;T _O3.TturtJE3.;RuQ;a3QNkFRpNpv;T aqpT3Q3Xq+l ]35NTturtl 1LLdh 9]H4jw ];*PtRpN;l ^hl HOrtqROl HhHhlZ;8tNqlEhJ DO Nqt R3Tt 3s 5QtR P3utTQ ;Ou 5QtR P3utTpOr pO _O3.Tturt[;O;rtPtON \+QNtPQl B3T5Pt m3saR3vttupOrQ 3s $F# #ONtRO;Np3O;Tl X58TpQqtu 8+ ^;5RtOvt (RT8;5P HQQ3vp;NtQl %Rttvtl XXh ig4kigel m44jh 9];4gw ];*PtRpN;l ^hJ CQtR [3utTpOr ;Ou atRQ3O;Tp*;Np3O 3s _O3.Tturt [;O;rtPtON \+QNtPQl #O FqtOl \h=hl [;r35T;Ql %h)h 7(uQh6 Hu;XN;8Tt ;Ou Hu;XNp2t $+XtRPtup; 7XXh mmgk mig6J #ut; %R35X a58TpQqpOrl m44gh 9]HZ4jw]3u;l Fhl HOrtqROl HhHhlZ;8tNql Ehl ];*PtRpN;l ^hJ CQpOr v3O2tRQ;Np3O;T ;rtONQ N3 Q5XX3RN Nqt ;u3XNp3O3sWO3.Tturt Qq;RpOr XR;vNpvtQh #ONtR;vNpOr .pNqF3PX5NtRQl (TQt2ptR 1g716JgdkMLl m44jh

25 9\vq4gw \vq.;R*l \hH v3ONt-N P3utT s3RXtRQ3O;TWO3.Tturt P;O;rtPtON ;XXTpv;Np3OQl #Oh ]3Nqk GtRrq3stRlEh]hl\vq5T*l \hl ^t;Wtl )hGh 7(uQh6J [3utTpOr ;Ou ]tNRpt2;T 3s F3ONt-Nh []F m44gl ^ZH#jLiel XXh 1M–jjl m44gh 9\[\4mw \tTTtOl Hhbhl [5RXq+l ]hl \q;.l _h^hJ $3. WO3.Tturt .3RWtRQ 5Qt Nqt .t8l #O aR3vttupOrQ 3s Nqt \#%F$# v3OstRtOvt 3O $5P;Os;vN3RQ pO v3PX5NpOrQ+QNtPQJ Fq;OrpOr35R .3RTulvq;OrpOr35RQtT2tQh [pOOt;X3TpQl [pOOtQ3N;l C\HJ HF[l m44mh 9\_4dw \qpP;*5l $hl _3pWtl \hJ _[mh4J G5QpOtQQ WO3.Tturt Qq;RpOr pO Nqt At8 mh4 ;rtl Ztv EtvqOpv;T b35RO;T m7m6Jg4kgil m44dh 9E_gLw Eqp8;5Nl bh Ahl _tTTt+l $h $hJ Eqt Q3vp;T XQ+vq3T3r+ 3s rR35XQI Zt. =3RWJ ApTt+l 1LgLh 9Bpo4gw Bpor;Ql &h Ghl GT3rrtRQ: t-XtvN;Np3OQ3sXRp2;v+ ;Ou ;vv35ON;8pTpN+J HO pOpNp;T Q5R2t+h b35RO;T 3s F3PX5NtRk[tup;Ntu F3PP5Opv;Np3Ol 147j6l ;RNpvTt 1ml m44gh

26 EnablingSocial MediaContent QualityAssurance usingSocial NetworkAnalysis

Maria-Amparo Sanmateu1,Matthias Trier2,Andreas Lienicke3,Andreas Rederer1 1 Deutsche Telekom Laboratories Convergent Servicesand Platform Deutsche-Telekom-Allee 7 64295 Darmstadt 2 TU Berlin Department forSystems Analysis Franklinstrasse28/29 10587 Berlin 3 T-Systems,Systems Integration Innovative Communication Services&Access Solutions GoslarerUfer35 10589 Berlin Germany [email protected], [email protected] [email protected],[email protected]

Abstract: User Generated Content(UGC) is one of thebases of Web2.0, itsquality controla criticalissue. This paperdescribes theworkdone at T-Laboratories to specify anddevelop an innovative adaptable QualityManagementSystemfor UGC leveragingranking,priorization and contentexclusionofUGC.The ContentQuality AssuranceSystem(CQASystem) is an innovative modular combinationofmultimediaminingaimingtoserve differenttypes of content platforms,especiallythose basedonUGC (i.e. online communities). Oneofthe pillar technologies used in oursystemisSocial NetworkAnalysis (SNA). SNA methods to calculate KPI’S between social connections andevaluated andrated contentobjects have been proved to provideamore complete andvaluedinsight forcontentcompliance, community controlling, andidentification of valuable usersand contents inside acommunity. SNAmetrics proposequalitativeusercentered analysis of thecommunity life cycleversustraditionalvolume-basedmeasures,enablingefficient community gardeningand community marketingwithnew innovativecontentmanagementuse cases.Our combinationofSNA methods with traditionalcontentminingtechnologiesproposes enhanced identification of relevantusers andcontent, focusedintervention(incentivation),social media relevanceengines,multimediacontentanalysis overcoming therealityofimmature technologies fori.e.UGC-video.Next steps will enhanceour CQA Systemfor major communities support, andextendthe conceptfor othercontentmanagementareas.

27 1Motivation

Web2.0 enables websites with increasedinteraction,dynamics andpersonalization [Ore05]. Usersengageinforming virtualidentities andincreasinglyestablishsocial properties likerelationships or acommunity.Groups of people activelycontributetothe so called user-generated content(UGC).Because of this aspectthe web2.0 is also sometimesbeing referred to with sloganslike“read-and-write-web” or “bring your own content”. Thecombinationoflastingcommunicationartifacts andcontent resultsin increasedperceptionofthe link betweencontents andauthors [ErK00].Thisalso implies that theauthoring person movesintofocus andcan be recognized [TrB09].

UGC brings novelchallenges forthe provider of awebsite. Nexttospecificlegal aspects,likecopyrightsorconformity with ethical values, thecontent’squalityis difficult to ensure in runtime. Anotherissue is to keep usersmotivated to producenew interesting contents.Thiscan be donevia incentives that build up reputationorfinancial benefits.Massesofratherunstructuredcontentsare furtherhardertostructure and classify. Thehealthofthe user community is thus difficulttomaintain.

Presentsolutions forsocial mediaqualityassurance do proposeincomplete, inadecuate solutions forthe traditionalcontentmanagementuse caseslike, compliance,community monitoring, relevance of content, identificationofpower users. Most of thetimemanual processesare proposed,orthe usageofvolumebased measures withoutqualificationor followupofthe social mediadynamics andlifecycle. Usageofisolated technologies cannotproposethe required demandsofSocial mediacontentmanagement[VV07] [Zo07].

Thereforeaconceptualand developing effort wasinitiated at T-Laboratories in orderto proposeainnovative,presentlyuniquesolutionofamodular combinationofmulti- medial mining aiming to servedifferent typesofcontent platforms.Contentproviders andbusinessownersshouldbeabletotakeadvantage of such asystemand useitfor examplefor abetter marketing.

2Working principle

To approach to abovechallenges of managing UGCwebsites, we proposethe software- supported conceptofUGC qualityassurance – UGCQA. It is implemented as asoftware server in thebackend of theprovider’s website or canbeaccessedvia data transfer services. Thefollowing sections will give ashort overviewabout thesoftware implementation andits benefits.

TheContentQualityAssurance System (CQA System)isamodular combinationof multi-medial mining aiming to servedifferent typesofcontentplatforms.Itenhances them with data modelingand reportingprocesses.

TheCQA System consists of 5basic buildingblocks:

28 29 Aseconddesignprinciple is high modularityand adaptabilityofthe software solution. Thesystemwas designed to easily integrate multiple software components,e.g. classifiers, withoutthe need forlarge scaleintegrationadoptions.Thisisespecially relevantfor themedia classifierswhere multiple externalvendorswill be tested andeasy integrationwillbeofkey importance.

3Social NetworkAnalysis (SNA)method

In UGC websites,providers arenot broadcastinginformation to isolated consumers, but UGC usersformadense networkofinterrelationships,which givesthempower through feedback that in turn provides strong motivations to participate. UGC usersare lobbying andorganizing aroundthe providerscontents.For qualityassurance,thisimpliesthat understandingthese networks andtheir emergenceand evolutionisakeyobjective.

Theappropriate method forsuchanenvironmentisSocialNetwork Analysis (SNA) [Wik09] [SNA08].SNA in UGC focusesonthe analysis of organizationalstructures betweenusers andcontent objects.Thisaugmentsthe objectoriented view of text analysis anduserprofilingand provides theanalyst with additionalinsight aboutthe health andprosperityofacommunity’sstructuralproperties,i.e.the groupformation, users’ attention, or themotivationtoproduce contentand to participate in commenting in avirtual people network.

An online community in an environmentofusergenerated contentestablishes the following wide arrayofimplicit andexplicit relationallinks,which lend themselves to analysis with SNAmeasures:

• Actors relate to theirmessages.

• Messagesare referencedbyother messages, e.g. comments, postings,orstories.

• Actors canconfirm explicit friendship relationships with otheractors.

• Actors canhaverelationships to otheractors because they read andcomment each others’ texts.

• Actors canhaverelationships if they jointlyproduce or consumeacontent (e.g. taggingapicture,writing adocument, buyingthe same product).

• Actors canhaverelationships when bothbelong adefined grouporparticipate in an event.

• Actors furthercan have similaritybased relationships,based on similar properties (usergroups).

30 In theprevioussectionwementionedthe componentbased structure of theUGCQA software approach.For themeasurement of social networkmetrics, aspecial adapted derivativeofCommetrix Social Network Intelligence software hasbeendeveloped [Tri08][TrB09].

Thenetwork intelligencesoftwareisconnected to thecentral CQASystemdata model andretrieves all necessarydataabout users, contents,and their links in ordertoproduce aseries of networkmetrics. Themetricsare then visualized in amonitoring cockpit. For thedomain of networkanalysisinthe contextofUGCQA,the threedomains (1)user relatedmeasures, (2)contentobjectrelatedmeasures, and(3) networklevel measures are important. Thefirst twodomains of user andcontent relatedmeasuresidentifyvaluable usersand contents,leadersofgroupsand users, whichare already heavilyinvested in the community.The metricsrankthese usersand contentobjects andtrace their importance over time to satisfythe needsofthe selected usecases, e.g. user retentionand community monitoring. Thethird domain provides measures forthe complete network. Examples arethe size andgrowthofanetwork, thenumber of centralactorsversus peripheral actors,the increase in thedensity of anetwork,orthe connectedness and clustering.

Some examplemeasurestoindicate theperspective of SNAare listed in thefollowing table.

Table1:Examples forSNA measuresinthe contextofUGCQA.

Content_Popularity (CP)ContentPopularity indicates theattention that usersina community dedicate to acontentobjectbyreferencing,linking, citing it in theircontributions (i.e.numberofvotes,comments, etc. linking, citing,referencing an article, posting,comment,media element,…). CP showshow one contribution triggers discussion activityand gets into thefocus of thecommunity. Content_Centrality (CC) Identifyingvaluablecontent. ContentCentralityindicates the numberofusers that reacted on acontentobject. Whereas CP is showingthe attentionintermsofactivity, CC is showingthe attentionintermsofsizeofcommunity that reacted to thecontent object. CC henceidentifies contents with prominence among many users. Which contents that gathered thelargestgroup of referencingusers? User_Centrality(UC) Identifyingvaluableusers andleaders of groups.User Centrality indicates theattention auserreceives from otherusers measured by thenumberofother userscitingorreferencing theobserved user andthus formingarelationship. By such references theuser gets prominentinhis community andmanyother usersgather aroundhim andobserve hisdoings. This measureisalso sometimes referredtoasUser Work Centralitytodifferentiate it from ausers social (friendship) centrality. Here,workreferstohis contributionofcontents. Which usersattracted thelargestgroup of otherinterested/related users?

31 User-Social-Centrality SimilartoUC(UC indicates theattention auserreceives from (USC) otherusers measured by thenumberofother userscitingor referencingthe observeduserand thus formingarelationship). Thedifference is that USC is not measured by citing relationships but by confirmedsocial contacts (numberoffriends) User-Social vs Work Incentivation -MotivatingCommunity Climate. This measure Centrality (USCUC) computes theratio of USC andUCtoshowifauser’s work centralityisrelated to hissocial contacts. Furthershows,ifhaving friends is generallyrewarding forgetting attentioninthe network (asthe averagecorrelationofUSC andUC). Is investmentin friendshipleading to centralitystatusand thus is beneficial fora user? Core-Periphery Incentivation -MotivatingCommunity Climate. This measure Stabilization/Fluctuation indicates thechangeinthe membersofthe core groupovertime (CPSF) by computingthe top20% andbottom20% central usersvia a subsequent computationafter user centrality(UC). Then the percentageofchangeinthese lists is calculatedasanindicator for fluctuationinthe center andinthe periphery of thenetwork.Are newmembers gettingintothe center? Is thecoreand periphery in constant motion?

Thefollowing twofigures help to introduce thenetwork aspects of user generated contentinmoredetail. Figure 2represents auserembeddedinalargernetwork of many users. This graphcan be used to explain theconcept of user centrality. Afirst simple centralitymeasure is counting therelative shareofcontacts of onenode.The most centralactor is simply theone with themostdirectcontacts. Anotherway to quantify the centralityofanode is closeness.Here, thedistanceofanode to all othernodesinthe networkismeasuredvia averageshortestpath length. In adigital networkthismeasure indicateshow fast or efficient an actorcan access thenetwork andhow likely it is,that informationreaches him. Athird common measurefor centralityisbetweenness.It represents thenumberofshortest pathsbetween pairsofnodes, which runthrough the observednode.Inane-mail networkthiscould be thepersonwho forwards important messages andthusisimportant forthe informationtransfer betweenpairsofactors.This canbeanimportantnetwork positionbut is also critical forinformationtransferina communicationsetting.

Figure 2: Networkvisualizationofusercentralityinanetwork.

32 This conceptofcentralitycan be applied to theUGC context. Centralusers aretightly embedded in anetwork of otherusers andtheir contributions.These contentnodescan be seen as lowrisky.Futurecontentrelatedtothese nodescould be published automaticallyorcould be higher rated.The automaticallytaggedimpermissibletextalso transfersacertain amount of this trustonother contentnodesaround it – otherusers, othercontent. Everyother contentrelatedtothiscan be tagged with thestatus “ominous” to indicate thelackofaawardedprioritrust.

Figure 3: Contentreputationderived from user reputation.

4Usage of SNAinthe CQAS

Thenetwork analysis of community data logs is used to generate visual insights and automated measures forstructuresofexistingvirtual communicationnetworks. The following Figure 4gives an overview aboutSNA analysis options forUGCQA. Figure 4A showsasubnetwork whereseveral actors referenceeachother andthereby form a network. Theshown networkisanego-network of oneuser. Theobservedego is the largernode located in thecenter.Edges representhis connectiontoother usersinthe network. Thenetwork graphshows that variousofego’s contacts aredensely connected viadirectlinks (upperright area). Othersinturnare only connected viaego (bottomleft area). Forthe participatingnodes acentralitymeasure is computed.Further,their activity (amountofmessagessentinthe observedperiod) andpopularity is analyzed and measured.These measures canfurther be computed only forcertain topiccategories (B) to seeifactors have reputation in definedtopics. Figure 4B showsthatinone sample topic, thereare disconnected smallnetworksofinteraction, whereasinasecond topical categorythere is adensely connected discussion with theobservedego actorlocated near thecenter.Finally, theevolvementofcentralityetc. canbemeasuredoverseveral time periods (C). This longitudinalanalysisshows howthe egonetwork of theobserved node evolvedovertimetobecomeadense graphofconnected users.

33 Figure 4: Overview about theSNA measurement visualizations forUGC QA.

5Benefits andConclusion

This paperintroducedthe conceptofquality assurancefor user generated content. We focusedonthe role of social networkanalysistostudy thereputationofusers andthe qualityofcontents.Itcan be concludedthatSNA algorithmsprovide an integrated view of therelationamong usersand their content. Their metrics canbeusedaskey performanceindicators (KPI)and canbecombinedwithother contentmining methodologies to yield,for example, amoreefficient report forranking of newly uploaded community problematic /non compliant contentobjects (media, images, text).

Theusage of the enhanced users’ metadata canalsoallowanalysisinsomefieldslike videoon-line communities whereatpresentimmature videocontentanalysis[IAI08] still fails to produce satisfactory results. Acomplementarypartofthe concepts introducedinthispaper relatestoother measures, e.g. text analysis to assesshot topics.

SNAcan also providealternative KPI’stothose traditionallyusedtomeasure and monitorcommunities, like forexample counting of page impressionsoruniquevisitors [WIK09].SNA KPI’s introduce sociological concepts to community monitoring and measures, hencecreatingamore user centricapproach. SNA couldtherefore be used to measureSocial activity andqualify reactions to publishedcontentwithout thenecessity of couplingtraditional volume metrics with sociological analysis doneusing online questionnaires [AGO08].

WorkdoneatT-Laboratories hasspecified aconcept combiningmultiple existing technologies in anew manner – forimprovedanalysisand exploitationofUGC data:

34  Generic database structurewithasetoftables to store SNA(Social Network Analysis)calculated data  KPI’S(KeyPerformance Indicators)based on contentanalysis(media, text)  Database importand thereforeclearseparationbetween CQAS andcustomer  Network analysis –tocalculateKPI’S betweensocial connections andevaluated andrated contentobjects (media analysis,textanalysis, near to suspectcontent providersetc).

Nextstepsofthe on-going work is to enhanceand scaleour already developedprototype to supportmajor communities,toextendthe existinguse casestowards othercontent management issues like relevance or recommendationcontent engines.

References

[Ore05] O’Reilly,T.:WhatIsWeb 2.0. 2005.URL:http://www.oreillynet. com/pub/a/oreilly/tim/news/ 2005/09/30/what-is-web-20.html [TrB09] Trier,M.; Bobrik,A.: Searchingand ExploringSocial Architectures in Digital Networks. IEEEInternet ComputingJournal.Mar/Apr,2009. [Tri08] Trier M.: TowardsDynamic Visualizationfor UnderstandingEvolution of Digital Communication Networks.InformationSystems Research,Vol.19Nr.3, 2008, S.335- 350. [ErK00]Erickson, T.;Kellogg, W.A.:Social Translucence: An Approach to Designing Systems that Mesh with Social Processes.InTransactions on Computer-Human Interaction. vol. 7, no.1,ACM Press, NewYork, 2000. S. 59-83; URL: http://www.research.ibm.com/SocialComputing/Papers/st_TOCHI.htm. [VV07] VanVeen N,JacksonP.Social Computing. TheRoleofTelco’s in User Generated Content. Forrester. StrategicalStudies.21August2007. [Zo07] Zoller E, Operator Strategies forUGC andsocial networking.OvumStudies,22February 2007. [WIK09] Social networkanalysis andMetrics (Measures)insocial networkanalysis URL: http://en.wikipedia.org/wiki/Social_networks [SNA08]SNA -Social NetworkAnalysis, ABriefIntroduction URL: http://www.orgnet.com/sna.html [WIK09] Unique visitor definitionand usage in onlinemarketing URL: http://en.wikipedia.org/wiki/Unique_visitors [AGO08]MeasurementMethods of ArbeitsgemeinschaftOnline Forschung (AGOF) URL: http://www.agof.de/methode.585.html [IAI08]Dr. Joachim Köhler. Fraunhofer IAIS,Institut IntelligenteAnalyse-und Informationssysteme: Semantische Verarbeitung vonYouTube Videos.Erste Ergebnisse und Konzeption zurAutomatischeAnalyse vonUGC Videos.. Berlin September2008.

35 Contributing andsocialization –biaxialsegmentation for usersgeneratingcontent

HendrikSend, DanielMichelis

InstituteofElectronicBusiness An-Institut derUniversitätder Künste Berlin Hardenbergstraße9A D-10623Berlin [email protected]

Anhalt University of AppliedScience StrenzfelderAllee 28 D-06406Bernburg [email protected]

Abstract: User generatedcontentisavaluable resource voluntarilyprovidedbya growingnumberofusers.Asonlineand traditionalbusinessesincreasinglyharness this resource theneed forastrategicway to manage usershas become apparent.

In this paper, we seek to establishamodelthatappliesprior research insights regardinguserbehaviour to user generatedcontent. Therefore, we developa strategicsegmentationofusers whocontributecontent. Usingabiaxialmodelput forwardbyKozinetsfor thesegmentationofconsumers andanordinal scalemodel by Li andBernoff, we proposeamodeltograsp boththe degree of social involvementand theintensity of contentcontributionbyusers.

1Introduction

Since theinceptionofthe Internet,its usehas largelybeen limitedtothe passive receptionofavailablecontent until theadventofprominent,participatoryplatforms like Wikipedia, YouTube, andFacebook,which have allowedonlineusers to build content resources collectively. Asubstantialnumberofwebsitesand services areemergingthat draw considerable valuefromusergenerated content. They arebecoming an economic phenomenon greatly affectingthe design of contemporarybusinessmodels andthe onlinemedia landscapeasawhole, as well as traditionalmedia andmarketing [WV07].

Giventhe astounding growth of user generatedcontent,businesseshavebegun exploring howtobestbenefit from theirdevelopmentand ways to exploitpossiblecompetitive advantages.Inspite of theseefforts,the dynamics of user contentare still poorly understood, whichhas deterredbusinessesfromtakingadvantage of itspotential economicpossibilities.

Forthose businessesthatalready engage activeonlineusers,theyfindthemselvesina

36 race to secure theirusers’patronage before thecompetition. Forinstance, thereare severalonlinemovie review communitiesinthe UnitedStates, whereusers make contributions rangingbroadly in nature andquality[Oz01].The economically successful moviecommunity is onethatbestcaterstothe demandsofits usersthrough theproducts andservicesitprovides.

Furthermore, traditionalbusinessesare discoveringthe valueofusers whodonot necessarily file formal complaints butactivelyintroduce newideas throughonline formats, as can be seen in thecaseofthe onlinecommunity myStarbucksIdea. To the same end, less well-known businesses likethe airlineJetBlue have startedcomparable programs.Itisforeseeable in thenear future that bothStarbucks andJetBlue will face competitorstothe online idea platform andhavetoformalizetheir commitment from theironlineusers.Therefore we need to developdiversified strategies to attract anddeal with theparticipatingonlineusers.

Practitioners andresearchersare nowtryingtounderstandthe broaderimplications of user generatedmedia contentand howtofacilitate thecontributionofusercontent.In response to theseconcerns,thispaper will putforward asegmentationmodelofusers whocontributecontenttoonlinecommunities.

Ouranalysisisstructuredasfollows:First,weexamine existingusersegmentations for onlinecommunities. Next, we identifythe most useful segmentationmodels andupdate them in ordertoapplythemfor user generatedcontent.Inconclusion, we consider the implications of oursegmentationmodelfor theory andmanagement.

2Theoreticalbackground

We define online communities as agroup of userswho communicate, interact and developrelationships in atechnology-supportedenvironment[LVL02].

User generatedcontent (UGC)isdefined by theOECDthrough threecriteria[WV07]: First,the work hastobemadeaccessibletoagroupofusers (e.g.anonlinecommunity). Secondly,the editorhas to make acreativeefforttoproduce theworkormake adaptations from existing material.Finally,the developmentofUGC hastotakeplace predominantlyoutside of thecreator’s professional realm.

Online communitiesdonot necessarily automatically implythe creationofUGC;suchis thecasewithinstant messaging communities. Nonetheless, UGCiscentral to the structureofmostonlinecommunities, as previously establisheddefinitions propose [HA99]:“Virtualcommunities arecomputer-mediated spaces wherethere is apotential foranintegration of contentand communicationwithanemphasisonmember-generated content.”Wetherefore define communitiesofUGC as online communitiesinwhich UGC playsasignificantrole. TheimportanceofonlinecommunitiesofUGC is exemplifiedbydata. Thesitesthatfallintothiscategory(Wikipedia, MySpace, Piczo, YouTubeand Bebo) exhibitaconsiderably higherfrequency of generalusage andrepeat visits than othersitesrankedinthe top50sitesinthe UnitedKingdom [WV07].

37 2.1Usersegmentationasautility

Innovationmanagementproposes awell-researched user segmentationfor the integrationofasegmentofsocalled“lead users” into thedevelopment of newservices andproducts[Hi86].Asleadusers nowadays also contribute informationtoonline communities, they can be identified andmoreeasilyintegratedintothe innovation processes[BB08].

Innovationmanagementismainlyinterestedinthe expectations andideas of certain users. Agrowing number of onlinerepositories of lexical articles, product reviewsand bookmarksrelyheavily on thequalityofUGC.Therefore,numerous researchershave analyzed thecharacteristicsand differencesofthe contributions made by varioususer groups.For example, Stein andHesshaveexaminedhighquality, featured articlesinthe Germanlanguagesectionofthe online encyclopaedia Wikipedia. They created asystem to gauge thequalityofaneditor’sworkand then assessed therelationshipbetween the editorofthe article andthe article’sproficiency [SH07].

Theabovementionedsegmentationanalysesusergroupsafter user contributions. In ordertosufficientlyaddressusergroupsand in advanceofthe long-term establishment of agiven online community we need to find asuitableusersegmentationthatwill facilitate thedesignofonline-communities.

2.2Early segmentations: Goal-directed andexperimental or Passionates, Pragmatics,and Phobics

An early approach in categorizingusertypes wasput forwardbyHoffmannand Novak. They distinguishedtwo kindsofusermotivation: goal-directed andexperiential[HN96]. Thedichotomous modelisbased upon an opposing conceptualisationofextrinsic and intrinsicusermotivation. Accordingtothismodel, intrinsically motivatedusers find reward fortheir online actioninexperientialusage like diversionand relaxation. Whereas extrinsically motivatedusers make useofthe Internet with aspecificgoalin mind such as gatheringinformation. This modelhas been disputed on thebasis that users areusually compelledbybothexperientialand goal-directed motives [KW99].

Subsequently,Rodgersand Cannonfurther scrutinized thesemotivations (aswellasthe actualusage of online media),concludingthatthere arethree clusters of webusers: “Passionates”, whoare motivatedbyexperientialmotives,“Pragmatics”,who aregoal- directed,asseen in theabovemodeland “Phobics”, whorarelyuse online communicationand,ifso, strictly forgoal-directed purposes [Sh02].

2.3Two factor modelbyKozinets

Forthe specificpurposeofonlinecommunitiesofconsumption, Kozinets hasdefined a typology of user groups.Kozinetsdefines communitiesofconsumptionas: “groupings [that] areimplicitlyand explicitlystructuredaround consumptionand marketing interests” [Ko99].For this purpose, Kozinets describestwo central measures that affect

38 the“formationoflasting identification” with thecommunity:the intensity of relationship within community andthe centralityofconsumptiontothe user.

When an Internet user firstvisitsanonlinecommunity he or shewill predominantly browse forinformationand will be unlikely to have affiliations with many otherusers. However, as theuservisitsthe community more frequently andbecomesacquainted with howitworks,heorshe will potentiallybegin to engage in discussion topics and eventually developstronger ties within thecommunity.Depending on one’sstatuswithin thecommunity,users make useofdifferent communicationtools. Thus,Kozinetsdefines thesocialtieswithinthe communityasafactor of primaryimportance[Ko99].The second significant factor in thecaseofconsumptioncommunities is howhighofa priority theconsumptionactivityistothe user.The more central theconsumption activityistothe user’s self-image,the more he or shewillvalue thecommunity.The self-centralityofconsumptionisnot independentofone’s social ties.

Illustration1:Usertypology forcommunitiesofconsumption[Ko99]

Illustration1showsfoursegmentsofusers on twoaxes. Theintensity of relationshipis representedbythe horizontal axis andthe centralityofconsumptionisrepresented by the vertical axis.

Kozinets typology hasbeen studiedinvarious research projects. Themainadvantage of such asimplifiedmodelisthatitcan be appliedtoseveral casesand researchersare able to developabody of in depthknowledge aboutthe propertiesofthe user segments.For instance, researchersinthe fieldofcustomerintegrationhaveadopted this Kozinets scheme to identify lead usersfor product developmentinseveral cases [FK06, Fu04].

Theabovementionedsegmentationisespecially useful in thefield of customer integrationbecauseitspecificallytargets communitiesofconsumption.However,for our interest in user actionwithinonlinecommunities, we proposeanadaptationofthis

39 Kozinets model, altering its focusonconsumptiontothe generationofusercontent.

Illustration2:Usertypologyfor UGC[LB08] Li andBernoff classify usersaccordingtotheir involvementincontent generationona profile ladder. "Each step on theladderrepresentsagroupofconsumers more involved [...] than thepreviousstep."[LB08].The toprung of theladderisoccupied by thegroup calledthe Creators, whopublishblogs,havetheir ownwebsites, upload self-created videos andaudio or write articles andstories.Below theCreator is theCritic,who posts ratings andreviews of productsorservices,commentsonblogposts,contributes to or edits online forums andWikipedia articles. Theusergroup thirdfromthe topis comprisedofCollectors, whouse RSSfeeds, addtagstowebsitesorphoto collections or vote forwebsites. TheJoiners maintain aprofile on asocialnetworkingsiteand also visitthese sites. TheSpectatorsare passive contributors whoread blogs, watchvideos from otherusers,listentopodcasts,and read forums andonlineratings.Lastly, the Inactives do nottakepartinany of theactivitiesmentionedabove[LB08].For each user segment, Li andBernoff developedstrategiestobuild andmaintainappropriate relationships accordingtoagivenorganisation’sobjectives.

It is reasonable to assume that social affiliations grow alongwiththe generalonline activitylevel of auser, though thereisnot necessarily an increaseincontent generation commensuratewithanincreaseinuser’ssocialaffiliations.[SGV04].Additionally, users maywrite many articlesorreviews withoutmaintaining aclose networkor, inversely, mayhavemanyaffiliations withoutcontributing content.

3Usergroup modelfor contentgeneration

As we endeavourtoadapt aprovenmodelofsegmentationfor communitiesofcontent generation, we have chosen theabovebiaxial modelbyKozinets, whichhas been used

40 by many otherresearch projectswithsuccessfulresults,althoughweupdate the consumptionactivityinKozinets’ modeltosuitour needs.

Thefirst factor in Kozinets’modelisthe centralityofconsumptionactivity. Sinceour work is focusedonthe generationofusercontent,wepropose theaccordantmeasure for consumptionactivityas: contentgeneration. We have seen that thedistributionof contentgenerationisnot even among theusers of online communities[LB08].Anumber of studiesidentifyample differencesamong theusers whoshowdifferent levels of contentcontribution. Themeasurement of UGCcan be complicated,since thesame contentmay be featured on multiple sites, registered usersmay be inactiveorhave duplicateaccountsand thedistinctionbetween user-created andother contentisdifficult to discern[WV07].However,researchersmainlyemploystraightforward methods of measurementsuchasthe number of edits on articles[SH07],numberofbookmarks [BK08] or number of ideas contributedtoagivencompetition.

Thesecond factor presentedbyKozinetsisthe intensityofsocialrelationships that a user has. This measureisanalogous to therelationalcapitalasconveyed by research on social capital. Because theresearch on social capitalhas informed muchofthe understanding of online communities,the intensityofsocialrelationships should be understood as formingapart of thenotionofsocialcapital. Social capitalisgenerally definedasanonmonetaryformofcapitalthatprovidesanindividualaccesstovaluable resources like jobs or information[Pu95].For this reason,socialcapitalcan foster the developmentofnew knowledge[NG98].Socialcapitalisalsorepresented structurally andfacilitatesthe agency of actors within this structure.

Social capitalhas threedimensionsthatcan be clearly distinguished. Thestructural dimensionofsocialcapitalcan be describedasthe tiesthatconnect actorswithinagiven network. Here, we canexamine more closelythe configuration, hierarchy, anddensity of thenetwork.Also, thereisthe relationaldimension or “relationalembeddedness” [NG98],which refers to thenatureofthe relationbetween oneactor andanother,within agiven network. It can describe howmuchinfluence usershaveupon others,ifthey shareafriendship or mutual respect forone anotherand what historybinds them together.Lastly, thecognitive dimensionpointstosharedcognitiveresources used in deriving meaningand making interpretations among thegroup.

Researchersoften utilizesimplifiedmodels of social capitalwhereas an affiliation betweenusers is definedbyadocumentedaction(e.g. having reciprocal email communication) on adichotomous scale(connected /not connected). Even thesemodels allowfor thecalculationofsocialcapitaland measures like centralityand theirlinkage to otherattributes like knowledgecontribution[WF05].

User generatedcontent is definedbythe fact that multiple usersworkoncontent together [WV07].Research in thefield of social capitaldemonstrates that peoplewho possess many tiestoothersalsohaveaccess to more sourcesofinformation, more ideas andmorehelpingeneral [Pu95].Therefore,UGC that springsforth from an environmentofabroadernetwork of communicationcan be of significantly higher qualitythanworkderived from scarce communicationsources [TW03].Exemplifying

41 this notionare thefrequentlyeditedfeatured articlesonWikipedia and, or ideas for innovations that areimprovedbycollectiveeffort in online communities.

Nowshiftingfocus from communities of consumptiontowards communitiesofcontent generationsuchasWikipedia,Amazonand Threadless, we combinebothmodels, identifyingcontent generationasthe principalfactor,which is equivalent to the centralityofconsumptionactivityfromIllustration1.Illustration3showsthe user typology forcommunitiesofUGC.

High Critics, Collectors Creators

Content generation Inactives, Joiners Spectators

Low

Low High Social capital

Illustration3:Proposedadopted user typologyfor communitiesofUGC Li andBernoff make adistinctionbetween Inactives andSpectatorswith regardsto contentgeneration. Yet,bothgroupsare describedasnot activelycontributingcontent to onlinecommunities or exhibiting social tiesinonlinecommunities. Therefore, we argue that from theperspectiveofcontent generationand social capital, thetwo categoriescan be seen in asingle, shared group. TheJoinerprofile is describedasvisiting social communitiesand only maintainingaprofile there. Howeverthe OECD definitioncalls foracombined effort to createUGC.So, we find considerable social capitaland low contentgenerationfor theJoiners,characterisingthemasbeing of high social capitaland weak contributorsofcontent.

Collectorscontributetocontent repositories throughvotingand tagging andthereby generating valueinonlinecommunities. They do notexplicitlymaintainsocial affiliations online andcan be seen as activecontributorsofcontent while still having low social capital. Finally, Li andBernoff differentiate between Criticsand Creators, butthe less activegroup,the Critics, still contributesregularly to online forums andwikis.Both groups hold high social capital throughtheir activeroleinonlinecommunitiesand because of this,wecombine thetwo groups in thefield of high contentgenerationand social capital.

Forour purposes,thiscombinedmodelisanimprovement upon Kozinets’modelasit hasbeen adaptedtothe fieldofUGC,offering more clarity by indicatingthe implicit components of social capitalthatLiand Bernoffincludedintheir mono-axial modelof

42 contentgeneration.

4Towards operationalisation

Future research will have to operationalisethismodelinorder to properlysituate authors on thematrix. However, theoperationalisationofsocialcapitalonasingle scaleposes a challenge becauseitisunderstood to consistofmultipledimensions [NG98] andlacksa commonlyagreed upon definition(Glaeser 2002). Nevertheless, current research has establishedasystem of measuringsocialcapital[DS05].

In this paper, we proposethree dimensions of social capital forconsideration. Firstly, structural capitaldescribes thepositionofanactor within agiven network. To determine thenetworksconfiguration, thecorresponding networktie betweentwo actorsmustbe defined. Forinstance, we canassumethe existenceofaconnectionbetween twoactors when they have demonstrated reciprocal communicationinthe past (e.g.questionand answer)via commentsoremail. In this case, thenecessary data couldbederived from logfileanalysis. Additionally, graphtheoryoffersseveral meansofmeasuringstructural capital, such as centrality, betweenness, closenessoreigenvector centrality. Waskoand Farajequatestructuralcapitalwith thedegreeofcentralityofindividuals in anetwork [WF05].

Secondly,relationalcapitaldescribes theintensity of relationsbetween variousactors. This intensitycan be seen in things such as mutuallysharedvalues, trust, andperceived reciprocity;all of whichare underlyingmeasuresthatcan be assessedbyLikertscale questionnaires. Lastly,cognitive capitalisdeterminedatanindividuallevel andcan be equatedwithanindividual’slevel of expertiseinregards to theonlinecommunity that they arememberof. Depending on thescope of application, expertisecan be self-rated or assumedtocorrelate with theperiodofmembershipwithin an onlinecommunity.

An assessmentofthe differentmeasures of social capital andthe combinationof structural,relationaland cognitivedimensions is beyond thescope of this paper. Many authorsargue that social capitalshouldnot be subsumed under asingleformof measurementbecause this wouldundermine theexplanatory powerofthe underlying constructs [FKS07].Still, Mathwick,Wiertzetal. successfully combinethree comparable measures (voluntarism,reciprocity, andsocialtrust)intoalatent social capitalconstruct by buildingthe linear sumoftheir threeconstituents[MWD08].This approach couldserve as apreliminary step towardsoperationalisationofthe social capitalmeasure.

In contrast,content contribution can be measured by defining acriterion forcontent and tallying thecorresponding contributions.Inthe caseofmessagesonbulletinboards, for example, eachmessage must be sorted into categoriestodistinguishbetween knowledge contributions andquestions or asimpleexpression of gratitude such as:“Thanks!” If certain contenttypes such as voteorideaare predefined,their totalnumbers can be made useof.

43 5Conclusion

Theever-increasing flow of informationonthe Internet poses an enormous challengeto researcherstryingtoidentifyand applysuper ordinate structures to understandonline communication. In this paperweintroduced apreliminary modelofusersegmentsthatis applicable to variouscommunitiesofUGC.

Concerning theory,there is,atpresent,nounifyingmodelthatdepicts thecore differences of user groups in respect to UGC.Fromatheoretical perspective, ourmodel serves as apoint of departurefor furtherresearch andimproved comparison between newinsightsthatare gained.For instance, research on themotivationofusers in thecase of Amazon.comreviews hasshown that extrinsically motivatedusers write significantly more reviewsthanothers. Yet, thereviews of theseusers aregenerally ratedasless helpful. Accordingtoour modelthese usersshoulddisplay lowerlevelsofsocialcapital, whichremains to be proven.

With regard to managerial application, thesegmentationofusergroupshas been a subject of concernsince long before theemergence of online communities. Ouruser modelprovidesafirststepinthe developmentofamanagerial tool foronline communitiesinorder to enhancethe amount of UGCbyallocatingservices to user groups.

We focusedonabasicmodelwithtwo factors. This constraint allowedustoapplythe modeltoabroadrange of online communities andhelpedtonarrowthe focusofour analysis.Admittedly, we hadtoexclude important criteriasuchasmotivationand demographicattributes,which aretoo multifaceted andcomplex foratwodimensional model.

In future research we will operationalisethe factorsofcontent generationand social capitaland examinethe depicted user groups formotivationalsources andattitudes. At that time we can begintodevelop strategies to directly addressthe motivationofeach user groupinaneffectivemanner.

6References

[BK08] Benbunan-Fich,R.; Koufaris,M.: Motivations andContributionBehaviour in Social BookmarkingSystems:AnEmpirical Investigation. ElectronicMarkets,18,22008; S. 150-160 [BB08] Bilgram, V.;Brem, A.;Voigt,K.-I.:User-centric Innovations in newProduct Development-systematic IdentificationofLead Usersharnessing interactiveand collaborativeOnline-Tool.InternationalJournal of InnovationManagement, 12, 32008; S. 419-458 [DS05] Davenport, E.;Snyder, H. W.:Managingsocialcapital. AnnualReviewofInformation Scienceand Technology, 39, 2005 ;S.517-550 [FK06] Fichter, K.;Beucker,S.: Wandelder Innovationsbedingungen in derInternetökonomie: erklärungsbedürftigePhänomene im Themenfeld Innovationund Internetökonomie. Fraunhofer-IRB-Verl. 2006.

44 [FKS07]Franke, N.;Keinz,P.; Schreier,M.: Complementingmasscustomizationtoolkits with user communities: Howpeer inputimprovescustomerself-design.InInternational Research Conference on NewProductDevelopment. 2007 ;S.546-559. [Fu04] Fuller, J.;Bartl, M.;Ernst,H.; Muhlbacher,H.: Community BasedInnovation—A Method to Utilizethe InnovativePotentialofOnlineCommunities. hiccs, 07, 1530-1605 2004;S. 70195c [HA99] Hagel, J.;Armstrong,A.G.: NetGainProfitimNetz;Märkte erobern mitvirtuellen Communitites. 1999. [HN96] Hoffman,D.L.; Novak, T. P.:Marketinginhypermediacomputer-mediated environments:Conceptual foundations.JournalofMarketing, 60, 3Jul 1996 ;S.50-68 [KW99] Korgaonkar,P.K.; Wolin,L.D.: AMultivariate Analysis of WebUsage.Journalof AdvertisingResearch,39,21999;S. 53-68 [Ko99] Kozinets,R.V.: E-tribalized marketing?:the strategicimplications of virtual communitiesofconsumption. European Management Journal, 17, 31999;S. 252-264 [LVL02]Lee, F. S. L.;Vogel,D.; Limayem, M.:Virtual community informatics: what we know andwhatweneed to know.InSystemSciences, 2002. HICSS. Proceedings of the35th AnnualHawaiiInternationalConferenceon. 2002 ;S.2863-2872. [LB08] Li,C.; Bernoff, J.:Groundswell :winning in aworld transformedbysocialtechnologies. HarvardBusinessPress. 2008. [MWD08]Mathwick, C.;Wiertz, C.;DeRuyter, K.:SocialCapital ProductioninaVirtualP3 Community.JournalofConsumer Research,34,62008;S. 832-849 [NG98] Nahapiet,J.; Ghoshal, S.:Socialcapital, intellectualcapital, andthe organizational advantage. AcademyofManagementReview, 23, 2Apr 1998 ;S.242-266 [Oz01] Ozer,M.: User segmentationofonlinemusic services usingfuzzy clustering.Omega, 29, 22001;S. 193-206 [Pu95] Putnam,R.D.: BowlingAlone:America'sDecliningSocialCapital. Journalof Democracy, 6, 11995;S. 65-78 [Sh02] Sheehan,K.B.: Of surfing, searching, andnewshounds: AtypologyofInternetusers' onlinesessions.JournalofAdvertising Research,42,5Sep-Oct 2002;S. 62-71 [SGV04] Sheizaf,R.; Gilad, R.;Vladimir, S.:De-LurkinginVirtual Communities: ASocial CommunicationNetwork Approach to Measuringthe EffectsofSocialand Cultural Capital. In Proceedings of theProceedings of the37thAnnualHawaiiInternational Conference on System Sciences HICSS'04 -Track 7-Volume 7. 2004 . [SH07] Stein, K.;Hess, C.:Doesitmatterwho contributes: astudy on featured articlesinthe german wikipedia. In Proceedings of theeighteenth conference on Hypertextand hypermedia.2007. [TW03] Teigland, R.;Wasko,M.M.: IntegratingKnowledgethroughInformationTrading: Examiningthe Relationshipbetween Boundary Spanning Communicationand Individual Performance. Decision Sciences,34,2Spring2003 2003 ;S.261 [Hi86] VonHippel, E.:Lead users: asource of novel productconcepts.ManagementScience, 32, 71986;S. 791-805 [WF05] Wasko, M. M.;Faraj,S.: WhyshouldIshare? ExaminingSocialCapitaland Knowledge ContributioninElectronicNetworksofPractice. MISQuarterly,29,12005;S. 35-57 [WV07] Wunsch-Vincent, S.:Participativeweb anduser-created content:web 2.0, wikisand social networking.OECD. 2007.

Session2

CooperativeInformationSystems

Sociallyenhanced reachability support within an IMS-community-based infrastructuresolution

ZuzanaKrifkaDobes,Karl-Heinz Lüke, AndreasRederer

Deutsche TelekomLaboratories, Innovation Development Ernst-Reuter-Platz 7, D-10587 Berlin,Germany

[email protected], [email protected],[email protected]

Abstract: At Deutsche TelekomLaboratoriesweare investigating how group communication can be enhanced with contextual information such as reachability. We demonstrate these methods using agroup communication scenario which is of interesttousfromauser and telco perspective. Thescenario includesfeatures such as messaging, presence and“click-to” style communication,including conferencing between friends andpersons with common interests. We summarize ouranalysisoftwo types of approaches to the next generation IMS platform capabilities, modular and layered approaches.Weinclude lessons learned from our prototypes efforts,which made useofIMS capabilities. We pointout someofthe technicalchallengesthat wereconfronted whilerealizing ashowcaseofuse cases involving groupscommunicationinvolving usersusingmobile devices and as well as users connected by webontheir homePCs.

1Introduction

Social network activities1 involveacombinationofad-hoc as wellasplanned communication among community members. To support this set of communication requirements –the underlyingarchitecture must be fairlyrobust. Thecommunity services architecture being developed by Deutsche Telekom Laboratoriesisbasedon IMS (IP Multimedia Subsystem [3GPP]) sinceitoffers afully-functional community infrastructure.The full-fledgedcommunity infrastructureand allcommunity end-user facing components that was developed is called CoSIMS (Community-based services on IP Multimedia Subsystem)and consistsofmiddlewareincluding aCoSIMS application enabler.Thissystemwas developedinternally, and has beenevaluated within anumber of user trials,and has beenfurther refinedinterms of theresults of these evaluations. Thebasic capabilities of this service includethe ability for community end-users to createand manage friendshipcircles within thecommunity, and to communicatewith one another,one-to-one or by group,bothoverthe weborovermobile [DO07, DO08]. The core middleware components which are usedincurrent prototypeefforts is called GEMS (GEneric IMS community Service). Figure 1shows theintegrated view of CoSIMSacrossweb,mobile and PC.

1 Themarket of mobile social communities will grow until 2012 [VI08].

49 Figure 1: Integrated community serviceenvironment Ingredients of thenext-generationcommunication scenarios which areheldinour current community service design aredescribedbellow. Each of these is supported to various degressbythe core APIs availablefromthe GEMSinfrastrucre.

• Managedidentity –Access to multiple social networksorcommunities. Theuser is able to usethese services withoutlogging in and outmultiple times. • Personalcommunication context –Users should be able to maintain their own personal blog,and socially-oriented message inbox. Additionally they can create theirown personal profileand configure their privacy settings andpreferences. • Community-based communicationcontext –Supporting 1-to-many or many-to- many communicationamong groupmemberswhich encourages group interaction.. • Social andspatial awareness –With thelocationinformation, theusers are able to view thepositionoftheir friends. Additionally, they can monitorthe activities of thegroupand get theirstatusinformationinthe group blog. • Shared community-based content spaces –Agroupisassociated with ablog. The blog is accessible to group members. Members can follow group activities through thegroup blog which is especially relevant when they are notableto participateindirect communication.

2Establishing acommunication context within online communities

Thefocus of ourcurrentresearch is on presence and reachability andmorespecifically we have focused on identifyingmeanswithin theservice to help users establish asense of contextwithin community communicationscenarios. Sincecommunities are assumed to be mobile, establishing contextisespecially complex.Our assumptionisthatwhen the context for communication is clear it helpstomaintain reachability and keepusers in touchwith eachother.Whenusers communicatewith others within theirsocialnetwork, the context is even more complex, andsocialreachability is achallenging but our service attempts to exposeanumberofresources to community groupsmembers to increase reachability. The context of communication includes not just the type of deviceorthe device environment of users, but also thecurrent user’s presence status, their location, andaswell as an indicationofthe activities within social communities that users are

50 engagedin. Knowing these aspectsrelated to the communication helps to encourage furthercommunicationwith community members. Thefollowing figure(cf. Figure2) shows thevarious influences on presence that ourservice designwill include.

Figure 2: Presence sources and reachability

3Social entities and related presence sources

We identify several kindsofsocial entities for handling ourgroup communication scenarios.Service users areabletocreate contacts, whichare required forparticipatory communication. Contactsmay be imported from varioussocialnetworks. If auserwants to have afullysynchronizedaccess to acontact that they introduceintothe service throughanimportordataentryprocess, the contact needstobemadeintoaregistered service user, which canbedoneonboththe mobile andweb though messaging containing active links. All users have access to acontact list of user entries which can include private, managed andopengroups. Groups are collections of serviceusers,and each usercan belong to multiple groups.Groups aremaintained and usedbyother service enablersrelated to presence management, groupmessaging,calendaring.Group managementfunctionality is made available to theother services viaanXDM enabler (cf. section5).

Each social entity typeisassociated with anumberofpresence aspects (status, location, reachability, blogging,and search) as well as various levelsofexposure to recipient social entities. Table bellow (cf. Table 1) shows thekinds of functionality available to an end-userwith respect to othercommunity social entities.

51 Reach- Status Location Blog Search ability On status, +Self R/W R/W R/W location, By status,time reachability +Trusted By status, time, RR ROnstatus Other membership By status,time, +Member of WR/W ROnstatus locationor PrivateGroup reachability +Memberof On status, By status, time, Managed WR/W R/W Location, locationor Group Reachability reachability By membership, +Contact List Calllog -Call logCall LogHistory location or reachability Table1:Presencerelated functionalityrelated to social entities In thetable aboveR(Read) indicated theability to view apresence aspect, and W (Write)meansthe ability to actually modify apresence aspect. Contacts that are not members of asocialgroup are associated with thebasecasepresence aspects which include theircommunicationhistory,which includestheir IM (InstantMessaging),SMS and voice call activity. For instance, ausercan publishanew status and make it available to other groupsorindividuals. They can also make theircurrentlocation availabletoothers,given that privacy agreements are exchangedbetweenthem to allow the release of location. If the users browse their community, theycan view the aggregated form of this informationfromwithin acustomized community application interface available on their mobiles and over their web. An extra interface is requiredto set up, configure and enter their status and reachability information.Anotherinterface is associated with viewing this informationfromapersonal or group perspective. We have designed theservice to make it complementary to social networkingapplications,but to differentiate fromthembyemphasizingreachability and personal productivity.

Managedgroups arecreatedbyspecificcommunity members,who sendout invitations to specific contacts to becomemembers of thegroup.Opengroupsare created by a community members, and thegroup is then listedasapart of an open group directory, and other community members canthenelect to join any of theopengroups.Private groups function likebuddy lists, in that membershave no awareness of beinginthe same group.

4Walkthrough of asocial reachability scenario

In this sectionasocial reachability scenario which was introduced in thebeginningof this paperwill be described in more detail. We includeall thefeatures which are the target of ourcommunity service, basedonGEM,plusextensions, whichrunsonboththe

52 web andmobile. We support in this examplemanaged groups,the ability to establish communication with group members, basicpresenceand reachability and group blogs. Let us followthrough with an exampleofthe scenario in action:

• Two friends, Harry and Hermione,still are decidingwhattodoonFridaynight, andstart gettingincontact with their friends.Harry hasconfirmed himselfas Hermione’s friend andgranted herpermission to be on hisfriends list, and visa versa. Harryentershis currentpresencestatus, which also includes additional information about ameeting he is currently attending, andthe best way to reach him, which is by IM on hisnotebook. • Hermione checksHarry’s presence status on her mobile phone andgoeshome and her status is automaticallyupdated to include the fact that she is at home and available overIM. • Anotherfriend, Arnold haspostedanitemonthe managed group presence blog, using his online community interface related to an event taking place on the weekend.There are pictures includedinthe group blog related directly to this eventsuggestionand alink to an eventinthe group calendar corresponding to the actual event. • Hermione respondswith an IM alert directly to Arnold’s presence posting. Hermione enters hersupport forArnold’s suggestionbyvotingfor it so that other users can see that this is somethingthatothers areinterested. • Later, Harry checks the group blog and findsout aboutthe eventthatisbeing considered for that evening,and sees that others have communicated their support for theevent. Harry makes agroup conference call over his mobile with Arnoldand Hermione,tosee who will be able to takeHermione to theevent,and basedonthispicks up Hermione and togethertheyvisit theselected event location. He updates thegroup blog to indicatetheyare now there. • Theevent sponsor, adance hall, publishes acoupon that Harryretrieves on his mobile as soon as he getsclose to theevent location. He is offeredadiscounted ticket forall of hisfriends.Harry publishes this coupononthe friendfeed. • As theparty is so much fun, Harryand Hermione both leaveacoupleof comments lateron, on their personalblogstoencourage others to join them next time.

These phaseswithin this scenariomay take placeamong many overlapping groups of friendsatthe sametime. In theinitial phase friendspropose interest in theideaof participating in events. Next events are proposed forthe evening agenda to friends. Friendscan respond with alternativeproposals.Comments to proposals can be made in theformofany groupbased communicationmodality including email, conferencing,or chat. The influence of an eventincreases when newcircles of friendsare involved.Next comes aphase where communication regarding alternative ideasorproposals take place. User content, suchasapicture is sharedwith members of thegroup.Inthisphase contactwith members of alternativefriendshipcircles or groupsmay also take place. Shared experiences and memories of theevent can be entered within thecommunity service, andsharedamong friends,with easy access to all related information and content, to anyone who has rightstoviewthe content.

53 In ordertosupportthe scenariomentionedabove,adetaileddescriptionabout the underlyingtechnical infrastructure, thelayered architectureand theenabler is givenin thefollowing section. Theservice is based on GEMS, which includes core elements of theCoSIMS middleware architecture which supportsprimarily groupcommunication and group management.

5Technical solutions

5.1Technical overview

Avariety of implementationprototypes of theunderlyingcommunity service infrastructure have been implemented.For instance, in oneprototypeand existingonline service was implemented with ourCoSIMS prototype. This meantintegrationof identities frombothservices. Theadvantage to theusers of theexistingonlineservice is that its users were able to engage in groupbasedcommunication. The extended capabilities of thenew service included conferencing andsomelimited participationof site activities overamobile. Thecommunity service was also implemented as astand- alone web-based prototype, with aWeb 2.0lookand feel with easy click-through functionality. Furthermore,several custom flash interfaces were built which communicate the core of CoSIMS.

Thelevel of integrationwithlegacy web sites dependsonthe commercial partners’ requirements. Theintegration can be rather deep,oritcan functionlike acompletely independentbut cooperative3rd partyrelationship. TheGUI can be customized and branded as requiredbythe service provider.

CoSIMS features are included in theinitial prototypes [DO07, DO08] andwere internally evaluated within ausertrial. This trial consisted of university students, which performed anumberofpredefined communitiy-related activities on their mobile devices. We alsoexamined usage logs,questionnaires, anddirect observationsabouthow the trial users integrated group communicationintotheir daytoday social lives. We gained insightintothe design of our service, which was followedupwith some redesignefforts of oursoftwaresolution.Certain aspectsofgroup communicationworked well, while others likegroup mobile conferencing neededfurther investigation.

Ourusersurveys also revealed thefact that requirements of older, established users were rather different from younger IP generationusers. The former demandedsimpler interfaces andrequirements and the latter demanded thelatest features related to presence and location-based services,and standard community features are expected. So as expected theservice providerhas to have agoodideaofwhich user population they are targeting.

54 5.2 Layeredarchitecture

Figure 3shows theessentialcomponentsthatare apart of thearchitecture of theexisting prototype.

Control&transportlayer. There are additionalkey functionalities that mark theIP Multimedia Subsystem as thefuture technologyinacomprehensive service and application oriented network.The IMSprovides easy andefficient ways to integrate different services, evenfrom third parties. Interactionsbetweendifferentvalue-added services are anticipated.

The particular techniques andmethodologiesthatare requiredtogainthese key functionalities are notnew,but theIMS provides thefirst major integration and the interactionofall keyfunctionalities. By defining logical entities that are connected to each other through standardized protocols, aplug-and-play architecture has beencreated that offers the possibility to physically place each functionatdifferentlocations andto assembleanIMS with functions from different vendors.

Figure 3: CoSIMSlayered architecture

55 5.3Enabler layer

TheIPMultimedia Subsystem[VW06, FO08] is defined from3GPPRelease 5 specifications as overlay architecture on topofthe 3GPP Packet Switched (PS)Core Networkfor theprovision of real timemulti-mediaservices. IMS can be used for any mobile access network technology, and it can also be used for handling fixedline access technology. This capability is specified in theNextGeneration Network reference architecturedefinitionbeing promoted by theEuropean Telecommunications Standards Institute (ETSI) Telecoms &Internet converged Services &Protocols forAdvanced Networks (TISPAN). Thecentral session controlprotocols are theSessionInitiation Protocol (SIP)[SC02]and Diameter [CA05].

Service Enabler Layer.The Service Enabler layer allows services to make useof telecommunications resources. For instanceaPush-to-talk over Cellular (PoC)-server [OP07] can be seen as an enabler that allows services to set up PoC-Sessions without having knowledge about howPoC really works.Theyare bound to thespecificnetwork technologies that they usetoprovide their functionality andthus do not offer independencefrom thecontrollayer but offerahigherabstractionsothatservices do not need to directlyaddress thenetworks but rathermay use specific enabler interfaces. For these reasons,the Web-services can be considered as amiddleware technology between this layer and theapplication-enabler layer. Enablers include presence, location, multimedia conference,multimedia messaging,community service enabler andPoC. Most of these enablers sit on the topofthe IMS interfaces andthusutilize allIMS offeredcontroland managementfunctions. On theother hand, service enablers such as multimediamessaging maymake use of theSMS-C [EM06,SM03]and MMS-C [EN06, PP06]tosendmultimediamessage to usersonthe legacy networks (GSM).

ApplicationEnabler Layer.The Application Enabler handles the support for user access, forusercontacts, forsharedcontent, locationinformation, andcommunicationoptions for users. This layer encapsulates service enabler functionalities and mediates the access to services.The ApplicationEnabler is capable of exploitingIMS capabilities that can help to carry outthe demands of thecommunity service [CA05, E146, E136].The enablerdeploys aweb service whichprovides direct access to the IMS enabling technologylayer [PM08].

6Providing functionality to service providers

We considerthe business scenariowhereby commercial applicationsneed to enhance theirexistingfunctionality with more group-based communicationfunctionality. In this case thefunctionality of ourextended IMS-based presence enabler can be provided withinatoolkit, accessible via aweb-servicesAPI.The value-add to the application provider is achieved through theirability to tap into theIMS enabled multimedia communicationcapabilities available throughaweb browser runningonadesktop or mobile client.

56 Our current approach is based on adistributed serviceinfrastructure where user informationisstoredinmanysystems.There must be ameanstofederate these user identities to allowthe service to takeadvantage of theCoSIMSToolkit. In thefirst step CoSIMSintroduces asimplefederation mechanism by exchanging auniqueservice specific userID. In this way auser maybeidentified to by aservice andinCoSIMS. This service ID is an anonymous ID,which carries no user specific data. Theuser himself hasfullcontroloverdata that might be transmitted fromthe service to CoSIMS suchascontact data. Services have no knowledge aboutauser’s federation status.

The Community Enabler Layer contains thecontrols which communicate with the platform –providedmainly as alibrary of JavaScriptobjects. Thecontrols will be downloaded to clientsand provide thecommunicationwith theCoSIMSplatform in a way that theCoSIMS features can be provided.Controls include those related to user access, personal settings, contact presence,userpresence, groupmanagement, group communication, group invitations,aswellasemail, messaging andinbox controls.

The Community Application consists of all provider-specificapplication level pieces of a service –but in ourshowcase remainsCoSIMSbranded. Acommunity application includesaregistration view, theservice portalviewand an exemplary community page viewwhich is the core community-based part of theservice. Each applicationwithin this layerimplementsacommunityservice, with aspecific goal and scope.These community applicationsuse theenablers fromapplicationenabler layer to trigger actions in the telecommunicationnetworks. They should notcommunicatedirectly with layer 2or1in order to be independent of anyspecificnetwork technology.

7Community convergence –integrating mobile devices

Communities have traditionally been split into 2differentsegments–themobile andthe onlinecommunities. Since thebackend of ourinitial community service offeringsis basedonanIMS infrastructure we decidedinour first client approach to provide access to thecommunity infrastructure viaapureIMS client. This community clientwas the basis foraextensive usability evaluation,aswell as for analysis of technical problemsof IMS-basedclients. Key findingsherewere limitedprocessing power andbattery lifetime of thechosenWindows CE /Windows Mobile platform.

Asecond client approach was also pursued–to solveboththe battery lifetime challenge andthe requirement to support multiple mobile platforms. Themobile clientbecame proprietary, exchanging presenceand notificationinformation onlywhennecessary and boundtothe IMS infrastructurevia aserver sideIMS proxytoexchange buddylist, presenceinformation, messages andIMS notifications.

Next we investigated the newestIMS frameworks togetherwith upcoming user interface packages. For the Symbian Series S60mobile we used Sun’sLWUIT2 –for user

2 See URLhttps://lwuit.dev.java.net/tutorial/index.html

57 interface support. Theresult leads to arapidlydeveloped and lightweight IMS-client which provided abroad base of deployablemobile platforms, nearlylikethe proprietary client approachabove.The prototypes also showedthe increasingpower and professionalismofnewly availabletoolkits andlibraries.

Theworld of mobiles is changing rapidly andtheir capabilities are improving continuously [BI08,CA07].Thisisparticularly the case for web browsers. Therefore we alsoconsideredaccessing the webuserinterface directly. Although there arevarious toolkits available forweb codedevelopment, thejavascriptcompatibilitytests showed increasingly good results.After user interfacemodifications theweb solutioncouldalso be shown on mobile browsers. Theusabilityfromauser perspectivecould of coursenot compete with the above mentionedmobile applications,because of limited browser speedand theneedtoadapt theweb UI especially for mobiles. One technique for improving theusability forthe users is to reducethe number of buttonsand information available on any single screen.

8Outlook

Providing end-users with direct communication over the web to members of theirsocial networkscan give providers of traditional services an advantage. Their offeringscan be mademorecompellingbybeing able to “plug-in”morefunctionality to their customers. Their customers are increasingly members of varioussocial networks,can be attracted to services which provide livecommunicationfunctionality. Theseenhancements can also extend thelifetime of thetraditional onlineofferings, as well as thescope of their existingcontent [VI08], andofferanatural community-driven feedbackpipetotheir service offering.Variousprojects at DeutscheTelekomhavelooked at group communication, using IMSand followupprojectsnow lookmorecloselyatimproving communicationbylooking at aspects of reachability and presence.

Community based services canbeexpandedtoinclude aunique combinationoffeatures, if theright setofenablers support themiddleware [ER05,OS07, OP07,IM06,OX07]. We have createdanumberofclientswhich are able to respond to various service requirements, comingfrom the mobile and online web-basedserviceareas[DO08, RE07]. We have implementedand evaluatedanumber of mobile solutions whichaccess anetwork-based communityplatform in various ways.Inthe market place, however, community providers are often forcedtoreact very quickly towards new emerging standards andtechnologies such as theBONDI3 initiative–and hypesaround newly launched devicessuchasthe Google Android G1 phone or theApple iPhone -aswellas theresponses of thedevelopercommunity. Community providers are certainly not able to follow all thesolution trends at every point, butrather should show flexibility in responding to market trends andshould notloosecontact to successfulcompetitive solutions.

3 See URLhttp://bondi.omtp.org/default.aspx

58 References

[BI08] BergInsight: Mobile Location Based Services, 2008 [CA09] Canalys: Website http://www.canalys.com/pr/2008/r2008112.htm, 2009 [CA07] Canalys: Smart Mobile Deviceand Navigation Trends 2007/2008,2007 [CA05] Calhoun,P.IETFRFC3588.Diameter Base Protocol. 2003 [DO08] Dobes, Z. K.;Dutkowski,S.; Schwaiger,R.: “Enablingthe Saturday NightSwarming Scenario using the CoSIMS IMS-based Community Infrastructure”, NextGeneration Networking Middleware,Samos, Greece, 2008. [DO07] Dobes, Z. K.;von Heynitz,D.; Rederer, A.; Sanmateu, A.; Schwaiger, R.:“IMSWeb ServiceControls for Extending ServiceFunctionalityinTraditional OnlineServices”, ICIN Proceedings2007 [E146]ETSIES202 391-14 “Open Service Access(OSA); ParlayXWeb Services; Part 14: Presence (Parlay X2)”,12/2006 [E136]ETSI ES 202 391-13 V1.2.1 “Open Service Access (OSA); Parlay XWeb Services; Part 13: Address List Management (Parlay X2)”, 12/2006 [EN06] ETSI ES 202 391-5 “Open Service Access (OSA); ParlayXWebServices; Part 5: Multimedia Messaging (Parlay X2)”, 12/2006 [EM06] ETSIES202 391-4 “Open ServiceAccess (OSA); Parlay XWeb Services; Part 2: Short Messaging (ParlayX2)”, 12/2006 [ER05] EnablerRelease Definition for IMS in OMA (Open Mobile Alliance),V10-20050809- A, August 2005 [FO08]FOKUSIMS Playground,http://www.open-ims.org [OP07] Open Mobile Alliance (OMA). EnablerRelease Definition for Push-to-talk over Cellular. Candidate Version 2.0–11 Dec 2007.2007 [OX07] Open MobileAlliance (OMA). XML DocumentManagementArchitecture. Candidate Version 2.0 –24Jul 2007. 2007 [OS07] Open Mobile Alliance (OMA). OMA Service Environment. Approved Version 1.0.4 – 01 Feb 2007,2007 [PP06]3GPP, TS 23.140 V6.14.0, “Multimedia Messaging Service (MMS); Functional description; Stage2”, June 2006 [PM08] Parlay X. http://www.parlay.org/en/specifications/pxws.asp, 2008 [RE07] Rederer,A.; Dobes, Z. K.; Sanmateu,A.: “Dynamic Services ConvergengeThrough EPS(Extended Presence Services) Services and Applications”, ICIN Proceedings 2006 [SC02] Schulzrinneetal.:IETF RFC 3261. SIP: Session Initiation Protocol, 2002 [SM03]Short Message Peer-to-Peer Protocol Specificationv5.0,19Feb 2003 [VI08] Visiongain: MobileSocialNetworking&Usergenerated Content, 2008 [VW06]Vingarzan,D.; Weik, P.; Magedanz, T.: “Development of an Open Source IMSCore for emerging IMStestbeds, academiaand beyond”, Journal for Mobile Multimedia, Rinton Press,2006

59 Towards Internet Communities to Help Improve the Wellbeing and Rehabilitation of Clinically Stable Chronic Patients

Leendert W. M. Wienhofen, Ingrid Svagård

SINTEF ICT Department of Software Engineering,Safetyand Security NO-7465Trondheim,Norway [email protected], [email protected]

Abstract: The EU projectNEXES introduces IT supportfor integrated healthcare, and attempts to prove throughclinical trials hownew technology can help improve healthcare. The web-based approach used in the Norwegian clinical trialon wellbeing for chronicallyill patients can be seen as averysmall health care community for the patients.Afterintroducing the NEXES project, the technical platform used forthe wellbeing case, and thecollaboration notation in BPMN, a specific case is described in detail and related to howthe collaboration between the patient and health carestakeholders can be seen as asmallgroup. Finally reflectionsare givenonhow theclinicaltrial couldbeextended by adding a ‘competition’ element using online communities in order to motivatethe patients to do their rehabilitation exercises,and whichtechnologystandard could be used for this approach.

1Introduction

Integrated carehas beenpinpointedasamajor goalfor many European projects, including theproject NEXES1 that we will reportoninthispaper. What characterizes integrated care is theneedtoput thepatients’needs in focusinthe face of multiple care providing organizations.This is howeverachallenge because collaboration among primary care, specialists, socialcaregivers,etc. often introduces an overhead into the care process that in the worst case can leave thepatient andhis/her needs outofthe loop.

1 http://nexeshealth.eu/

60 This is in particularthe case for chronically illpatientswho oftenmove from one institutiontoanother, often frequently. In addition,haltingcollaborationbetween caregivers is costly as it results in suboptimal disease follow-up in theprimary care services andhence increases the risk of costly hospitalreadmissions. The overall objectiveofNEXES is to improvequality of care and reduce costsbyimproving collaboration and optimizing thework-share betweenprimary and secondary care services.

In theNEXES project we address 4differentprogrammes, one of which is entitled “Well being and rehabilitation”.Thisspecific programme promotes early diagnosis andhealthy life-stylesofclinicallystablechronic patients, enhancing their self-management and improving compliance with prescribedtreatments.The principal componentsare physical activity andcognitive aspects.

In short,the wellbeing program includes:

• Instruction/tutorialsondisease managementand physical exercises.

• Physical “work-out” /exercise sessions in groups twice aweek

• Physical “work-out”/exercise program to be carried outonhis own, twicea week.

In Norway patientshavethe legal righttoa(paper-based) personalplanfor long term follow-up by coordinated (public) health services, this is called “individuellplan”, an individual plan. Theplancontains an overview of thecontact informationofkey stakeholders,the actions to take andthe plan-owner’s evaluationofthese actions. In NEXES we investigatehow this individual plan can be supported by IT toolsand how these openthe potential to become CollaborativeCare Plans (CCP).

In thenextsectionwedescribehow an online tool helps thepatient to bettercope with thedisease by creatingand followinganindividual plan.The sectionafter that gives a general introductiontohow BusinessProcess ModelingNotation2 (BPMN) is used to describe thecareprocess in ordertoidentifycollaborationbetween different professional stakeholders andthe patient.Thisisfollowedbyacase description andthenan impression on howacompetitive–health enhancing- element can be obtained when sharing informationonline with people that are undergoingsimilar treatment.

2 http://www.bpmn.org/

61 2Technical platform

In NEXES, we use theITtoolSamPro, developedbyVisma UniqueAS, to containthe CCP mentioned in theintroduction. In [Wa05],acase description is provided on howthe architectural description frameworkMAFIIA(Model-based Architecture description Frameworkfor InformationIntegration Abstraction) is applied to SamPro. Other than this publication,not much scientific informationisavailable on SamPro duetothe specialised natureofthe tool. Belowwedescribe themainpointsofthe tool,relevantto the case described in section 3.

SamProisaweb-basedprogram, giving theplan-owner (the patient)fullcontrol overhis ownplan. On theother side,municipal health servicescan followmultiple plans simultaneously,ofcourse,given that the patient has assignedaccess rights to them as there are very strict regulations as to who is allowed to view theplan.

The CCP is the plan-owner’s personal tool forfollowinguphis own treatment. In cooperation with hiscoordinator stakeholders and actionsrelated to the plan canbe registeredbythe userhimself. Eachactioncan be evaluated by theuser.Itisuptothe user of theCCP to register and invite stakeholders, which leadstoamore active participation in hisown situationand plan forfollow-up.

The plan-owner can alsoinclude apersonal social network in thefollow-up plan,for examplebyintroducing next of kinwhich then canexpress theirsupportbothonlineand by visiting thepatientmoreregularly.

To comply with Norwegian legislationand regulation,the clients arethe owners of the content and manage the access to theirown client information. Theclient are also users andtakepartinthe discussion upon ownhealth situation.

Each plan involves at theminimum one servicerecipient and one coordinator, due to the law-regulation.Coordinatorsare most often nurses or physiotherapists, rarely psychologists or doctors. Coordinatorsbelongmainlytoprimary health care services.

62 TheSamProplatform provides thefollowing services:

• CCP including mapping, action, goals and evaluation functionality is themost central service. Important informationinrelationtothe CCPisthe objectives of the planand evaluation of the client at different stages.

• CSCW –supporting sharing of information, messaging (internal e-mail),instant messagingand discussion board/blog functionality available in next version. Messages maybealerted by mobile phone SMS functionality

• Agenda–planning of client.

• Profile management –management of users is basedonaroles and access rights related to theroles.Caregiverscan have different rolesrelated to different clients.

• Patient managementmodule–it supportssimplemanagementofthe client profile.(Thepatient managementmoduleisnot connected to theelectronic patientrecordsatthe hospital or in municipalities.)

• Statistics:Statistical data canbeextracted,suchasamountofplans operativeor not in themunicipality/organization, level of use, etc.

• Logging: Available forall participants to see changes in planning activities done by whom andwhen, visitors to theplan(no changes made, but informationreceived).

2.1Mapping theinteractions

As indicated in theintroduction, many interactions betweenthe patient and thedifferent stakeholders take place. In order to understand andexternalize existing chronic care processes we have employed BPMN.Pleaserefer to [FS09] fordetails on whythis standard was chosen andthe lessons learned while creatingcare process diagramsin BPMN. An example of theNorwegian wellbeing care process describedaccording to the BPMN notation is foundinfigure1.Inshort,itshows that thepatientischecked for participation qualificationinthe clinical trial, andincase thepatientisfoundtobe suitable, theprocess is described from thepatientsand care givers view.The following sectiongives an overview over howthe patientand care services collaborateinmore detail (thecase is based on theinteractionsdescribedinfigure1)and how this is supported by theSamProtool introduced in theprevioussection.

63 Figure 1: Example of thewellbeing processrepresented in BPMN.

64 3Well being and rehabilitation programme –acase

In this section,acase fromthe well beingand rehabilitationprogrammeispresented as a storyboard. Theintention of storyboards in NEXES is to present as much as possible of theinteraction between thedifferent stakeholders andhow technologycan support these interactions. Thestoryboardbelowdescribesthe enrolmentofapatientcalled Rolf to the wellbeing programme; it startsatthe General Practitioners (GP) office:

The patient, Rolf,isgiven abloodpressure monitorfor home measurementsand apulse oximeterand instructed on howtouse it. He is also askedtoweigh himselfregularly. Dr. OlsenthenasksRolfwhether he has aPCwith abroadband connection and is familiar with using it. As Rolf confirmsthis, Rolf is told that Dr. Olsen wants to follow up on Rolf on-linevia theweb-basedcollaboration tool SamPro, and that he will contact him approximatelyevery 2or3weeks to checkonhis progress.Hewillreceive detailed training on how to usethe toolfrom theNEXES programmanager in thenextfew days.

Afew days after thevisit to theGP, Rolf is contacted by theNEXES program manager for detailed instructions on the exercise program and the follow-up he will receive. She teaches him how to use theCollaborative Care Plan(CCP) shehas created for him with SamPro.The CCP will be Rolfs own toolfor writingadaily journal, bi-weekly registration of thephysiological datameasurements, and bi-weeklyanswers to a checklist regarding hishealthand well-being.The registrations will help Rolf himself assess hisown health and help hisprogress, butitwill also help theGPtoprovide specialist advice regarding his disease,and theexercise teamleader to adjust his training program.Inessence, theCCP will function as acommunicationtoolfor all thepersons that are involved in thecareprogram for Rolf;the GP, theNEXES programmanager and the exercise teamleader. Averytinyprivate internet community indeed,though Rolf is instructedhowever, that he has full controlonwho is allowed to accessthe different parts of theplan,and see hispersonal data. This also meansthatRolfcan invite relatives into theprogramme.

In additiontocommunicationvia theCCP, Rolf will receiveweekly SMS’es with hints, tipsand reminders regarding hisexerciseplanand overalldisease management. The reminders are generated automatically, while more detailed informationsuchasthe hints andtipsare created by one of the healthcarestakeholders followingRolf’s progress.

Three weeksintothe program Rolf’s attitude to theprogram is positive, butabit ambiguous. He likes thegroup exercises,and he findsthe weeklyeducational session very motivatingand useful.Heistaughtabout howtocopewith thedisease, thepurpose androleofphysical exercise andinstructionsonexercises he can carryout on hisown. However he hasnomotivation whatsoever to exercise on his own. The SMS’es he receives every week sometimes helps him“remember” to do the exercises,but not always. Still, he duly follows up andanswers thequestionnaires, and carries outthe measurements. Afterthree weeks, theGPsends himamessage viathe CCP,tellinghim that he thinksheisdoing fine,and encouraginghim to keep up thegoodwork.

65 The exercise teamleader also checksRolf’sdata. Shelogs on to theCCP from her office andgetsanoverview over allthe patientscareplans that she is in charge of. Herweekly routineistoscan through theregistered data foreachone,and if shefinds deviations from expected values, shewillcall theperson by phone, or shewill leaveamessagein theCCP,for examplewith afollow-up question. Seeing that Rolf is having troublein doing hisown exercises,she decides to callhim thenext morning.

The case describedabove is an actual care programme description of theNEXES project. This specific case will runonatarget group andacontrol group,starting September2009. In thenextsection, we discuss additional technologythat can enhance this programme, though notethatitisnot part of theactual NEXES programme.

4Towards internet communities to help improve the wellbeingand rehabilitationofclinically stable chronicpatients

In the caseabove,anawareness aspect is very apparent.Knowing what information to provide to whom and whichinformation to receive enables the patient to be motivated to do theexercises andtherewith improvethe rehabilitation.Inasense,thistypeof awareness is similartowhat is being describedin[SR07], where software development teams were followedinorder to identify theawareness networks. In our case, to alarge extent, theawarenessnetwork is definedbythe patient following theCCP,specialised care giving group. Though,the patient caninclude others than just the stakeholders definedinthe CCP, in principle it is completely up to thepatientwhom to include, though,asthe SamProtool is not very open (due to legislation),somepatientsmight find it usefultoalternatives. By expanding theawareness network with others than thegiven stakeholders, thepatient is able to createhis owncarecommunity.

In [Ey08] we see that newweb technologyenables and facilitates 1) social networking, 2) participation,3)apomediation, 4) openness, and5)collaboration,within andbetween these user groups.Personal health records (PHR)suchasGoogle Health3 or Microsoft's HealthVault4 are two alternatives [to SamPro] wherethe patientcan register and publish hisPHR.The paper [Go08] introduces both systemsand howGoogleand Microsoft implementsecurity and privacy systems in order to gain thetrust of theuser.

3 https://www.google.com/health;“With Google Health,you can storeand manage all your health information in one place. In Google Health, you can also chose to share your medical records with family members, friends, doctorsoranyone else in your care networkinorder to coordinate your care. By creating and opening up an account, youhave the possibility to let other patients find youand create aprivatecarenetwork.” 4 http://www.healthvault.com/;“HealthVault offers you away to storehealth information from many sources in one location, so that it’salwaysorganized and availabletoyou online.HealthVault is working with doctors, hospitals, employers,pharmacies, insurance providersand manufacturersofhealth devices –blood pressure monitors, heartratemonitorsand more –tomakeiteasyfor youtoadd information electronically to your HealthVault record. With amorecomplete pictureofyour family’shealth, youcan workwith your healthcare professionals, and with all the Web sites that connect with HealthVault, to make more informeddecisions.”

66 Both systemscan be used to enter informationrelevantfor the wellbeing programme. Whenother patients with asimilarprogrammeare registered, they canshare their experiences, exchangeideas,motivate each otherand even‘compete’ in their rehabilitationprocess. We suggest that thedropping motivationtoexercise on their own –as describedincase in theprevious section- can be triggered by introducingsomesort of ‘competition’element. Thetypeofcompetitionelement enabled by online communities is well shown by Nike’s Men vs.Woman running challenge5.Eventhough this running competitionisnot ascientifically provenmethodfor verifyinganything,it doesshowthatitcan triggerpeople to ahealthier lifestyle by introducing an elementof competition.

TheContinua Health Alliance (CHA) hasproposed areference architecture [Ca07] (see figure 2) forafullend-to-endapproach to technology supportfor healthcare. The reference architecture itself is rathergeneric, thoughthe focusofthe CHAistoprovide interoperable wireless interfaces [Sc07] betweenhealth-related devices. The data emerging from these cancaterfor thehealthcare ecosystem as describedinthe paper.

Figure 2: TheContinua End-to-Endreference architecture(copied from [Ca07])

5 www.nikeplus.com

67 In ourwellbeing case, most informationistobeplotted into SamProbyhand,which in itself couldbedemotivating. In order to ease theproposed competitionelement (at least in atechnical sense), we agree with CHA that standardized wireless transfer of data helpstoenablethisasthe user doesnot need to worry aboutcables or interoperability. Workoutdataisuploadedautomatically andthe results can (ultimately) be automatically compared, giving back the motivation to actually do the exercises also whenbeing on theirown.

5Conclusion and further work

In this paperwehaveshown theNEXES approach to awellbeingand rehabilitationcase and we have proposed ideas which can helpenableabetterprogramme once technology is mature enough.

Theactual wellbeingand rehabilitationcase will be one of therandomizedclinical trials withinthe project, andwill be started in September. Thetrial will runfor 9months and afterthisthe useofthe tool will be evaluated.The competitionconceptinorder to prevent demotivationwhile patients are working out on themselves in their homescan not be part of this studyastechnologyisnot yet mature enough.

6Acknowledgement

Theworkreportedinthispaper is aresultofour involvement in NEXES. Partsofthe work arefundedbythe European Commission (Grant AgreementNo225025).

7References

[Ca07] Carroll, R. Cnossen, R. Schnell, M. Simons, D. ,Continua: An Interoperable Personal HealthcareEcosystem.IEEE Pervasive Computing,Publication Date:Oct.-Dec. 2007. Volume: 6, Issue: 4, pages: 90-94 [Ey08] Eysenbach,Gunther, Medicine 2.0: Social Networking, Collaboration, Participation, Apomediation, and Openness, JMed Internet Res. 2008 Jul–Sep; 10(3): e22. Published online2008 August 25.doi: 10.2196/jmir.1030. [FS09] Babak A. Farshchian,Ingrid Svagård, Usingbusiness process modelling to model integrated careprocesses: Experiences from aEuropean project. To be presented at: IWAAL’09 1st International Workshop of Ambient Assisted Living 2009 (IWAAL'09) 10th -12thJune,2009 -Salamanca, Spain [Go08] GregGoth, "Powertothe Patients," IEEE Distributed SystemsOnline, vol. 9, no. 5, 2008, art. no. 0805-o5001. [Sc07] L. Schmitt, T. Falck, F. Wartena, and D. Simons, Towards Plug-and-Play Interoperability for Wireless Personal Telehealth Systems, 4thInternational Workshop on Wearable andImplantable Body Sensor Networks (BSN2007), Springer Berlin Heidelberg, ISSN 1680-0737 Volume 13, 2007,Pages 257-263

68 [SR07] Souza, C.R.B.d., and Redmiles, D. "TheAwareness Network: To Whom Should I DisplayMyActions?And, Whose Actions Should IMonitor?,"ECSCW’07: Proceedingsofthe Tenth European ConferenceonComputer Supported Cooperative Work, Springer, Limerick,Ireland, 2007,pp. 99-117. [Wa05] Ståle Walderhaug, Erlend Stav,SteinLTomassen, Lillian Røstadand Nils Brede Moe, “MAFIIA –anArchitecturalDescription Framework: Experience from the HealthCare Domain,” book chapterinInteroperabilityofEnterpriseSoftware andApplications, D. Konstantas, J.-P.Bourrières, M. Léonard and NBoudjlida (Eds.). Geneva,Switzerland: Springer Verlag,2005, pp.43-54.

69 Bi-directional Distribution of eLearning Content for Cross-technology Learning Communities

Raphael Zender,Enrico Dressler,UlrikeLucke, and Djamshid Tavangarian Chair of Computer Architecture, University of Rostock ﬁ[email protected]

Abstract: This article describes the use of aservice-oriented architecture to bridge the gapbetween different eLearning types and tools. The basic concept is abi-directional distribution of web services provided by different eLearning environments. This is exemplarily veriﬁed by acombination of lectures that are held in amodern media supported lecture room with tools of the computer-aided face-to-face learning paradigm and within avirtual learning environment (Second Life). The newservice-based approach allows aﬂexible and systematic coupling of virtual and face-to-face teaching and learning to across-technology learning community.

1eLearning Scenarios and Approaches fortheir Integration

Today, computer-aided face-to-face learning is the most popular teaching and learning type. This includes the replacement as well as the extension of spoken and written content by digital media (likepowerpoint slides, simulations, demonstrations, etc.). These media can than easily be made available to learners in order to support their wrap-up phase. Therefor,the lecturer is mostly required to upload his slideshowand lecture notes manually into aweb-based eLearning platform in an asynchronous manner. As another example, online or virtual learning simulates real-life educational processes and enriches them with speciﬁc communication and collaboration elements where applicable (e.g. chats, forums, or shared whiteboards). eLearning platforms are often used to implement these processes, for instance for teleacademies. In this context, 3D virtual worlds hold asigniﬁcant innovation potential for eLearning as theysimulate the perceived social presence of alecturer and learners [COT07]. The commmunication mechanisms of the platform are synchronous and the individual learning processes takeplace at the same time. However, these environments are decoupled from real-life scenarios. In particular in virtual 3D worlds, virtual learning is acomparative emerging paradigm. Nevertheless, the number of education activities is increasing. Forinstance, the virtual world Second Life [Lin09b] has around 50 registered education events per day.The reasons for the success are the advantages overother solutions likechats, instant messaging and classical learning platforms in terms of communication and felt presence. Forinstance, IBM’sconference space [Lin09a] shows that virtual worlds really encourage people to so- cialize with each other.The virtual event participants truly felt as if theyhad attended

70 real-time events, interacting with others and carrying home practical information. This comes at one fifth the cost of atraditional conference and without asingle case of jet lag. Blended learning is adidactical approach to link between virtual and face-to-face learning. The learning process is pre-divided into phases, each with its ownorganization. Face- to-face learning phases are basically used for aguided lecture preparation and wrap-up. These phases are mixed with intensive eLearning phases for more informal scenarios. The givendivision into synchronous and asynchronous processes makes acompromise between technical possibilities and pedagogical needs. However, amethodical approach for aseamless linking between tools for virtual and face- to-face learning is desirable for learners and feasible from atechnical point of view. The users in this approach (lecturer and learner) can decide on specific modalities in an ad-hoc manner.This allows for instance aseamless combination of synchronous scenarios during the lecture (for the interaction between lecturer and learner) and asynchronous scenarios before and after (individual or collaborative preparation and wrap-up). The prerequisite is aflexible combination of used tools, platforms, and the corresponding processes. Fur- thermore, the approach should be applicable for scenarios involving different educational providers. This enhances the learning comfort, increases the scope and quality of alecture, and advances mobility and equality of opportunities for learners and lecturers. Along with the current evolution of the Webthere is another issue to concern. Learners are members of manydifferent communities [Gru07]. This includes learning communities (built by different eLearning platforms) as well as social communities (e.g. Flickr, YouTube, and Facebook). It is to be assumed that learners prefer learning material that is provided in their familiar communities/platforms. Thus, the required familiarization effort is lower than with dedicated newsystems. Hence, the content provided on the basis of agiven methodical approach (e.g. recorded lectures, slideshows, and relevant literature) should be applicable for the integration in different platforms and environments. The goal of our work wastosystematically interconnect classroom and virtual learning in order to provide ahigher levelofindividuality and flexibility to the user –not only in terms of a“3D remote control”, butasageneralized architecture for flexible, bi-directional interchange of media between different environments (likeclassroom, media lab, learning platform, virtual world). The systematic use of the bi-directional combination is promising ahuge benefit for electronic learning and enriches the experiences of learners and probably also those of teachers.

2Related Work

Computer-aided face-to-face learning as well as virtual learning have already been performed by manyeducational institutions. In particular from virtual worlds awide spectrum of innovative learning scenarios [COT07][Gol07][RH08] recently emerged. How- ever,there’sjust fewwork on the systematic, seamless fusion of face-to-face and virtual learning.

71 Some approaches, for instance the QuickWorlds Project [JMLL00], deal with virtual learning material (e.g. virtual 3D models), that have been integrated into face-to-face lectures through special equipment. Unfortunately,all these approaches are unidirectional, because the material comes always from the virtual world while learners interact from real class- rooms. An input of face-to-face learning material to enrich virtual lectures has not been taken into account. Another interesting project is Sloodle [KL06]. Sloodle integrates multi-user virtual environments such as Second Life with eLearning platforms such as Moodle. However, this project provides just an interface for one virtual environment to control another virtual environment. Sloodle does not consider face-to-face lectures, butthese are essential components in our understanding of immersive learning. The required integration of face-to-face lectures goes beyond the simple use of an interface likethe Second Life client. virtual participants should be able to consume more face-to-face content than just the voice of the lecturer.Therefore, immersive learning requires asystematic architecture that integrates additional media sources as the live lecture slides and avideo stream of the lecturer (for instance to watch live experiments). In particular this direction (face-to-face to virtuality) is where related works lack.

3Using the Service Oriented Architecturefor eLearning

In principle, different basic architectural models (e.g. Client/Server) can be used to connect different learning instances. An assessment on these architectures leds to the broker architecture (the fundamental model behind the popular Service-Oriented Architecture, SOA) as best suitable for the proposed fusion. It has the best scalability,fault tolerance, and performance, as fewbottle necks as possible, and avoids apriori knowledge on any eLearning instance. As figure 1illustrates, the broker architecture separates between instances by introduction of abroker,which dynamically determines the service of arespec- tive provider (e.g. alecture hall) that is best suited to fullfil the request of aconsumer (e.g. avirtual world) on the basis of service descriptions (e.g. video streaming of agiven lecture) registered by the provider. In an educational application scenario there are some additional points that alloworeven require the use of SOA. First of all, there is anumber of established network addresses that are known to all clients and servers (likelearning platforms); theycan be used as bro- kers. Moreover, the contentofalecture typically is not security-relevant, which simplifies the implementation and practical use of aprototype. (Nevertheless, across-institutional scenario requires basic services for authentication and accounting.) Another aspect is the large number of potential users and services that demands ascalability and agility im- possibly provided by conventional models (likepoint-to-point connection of clients and servers). Finally,astrong condition in educational scenarios is the acceptance and effective learning outcome by the users; not only those with aless technical background. This requires atransparent and easy-to-handle system architecture as well as an intuitive and satisfying client-side interface. The prototypical implementation of our system will be described in the following.

72 Figure 1: Aservice-oriented architecture can be used for the realization of acombined face-to-face and virtual lecture

4System Architectureand Prototypical Implementation

4.1 Typical Application Scenario

Advanced eLearning settings often use dedicated environments with special equipment as for instance the media lab shown in Figure 2. These provide alot of features besides the conventional presentation of slides. The available devices provide functions as for instance:

• Input and output of audio and video data (e.g. provided by cameras, projectors, microphones, speakers, personal computers, and laptops) • Signal distribution control (by acrossbar switch and awireless media control panel) • Taping and playback of audio and video data • Setup of avideo conference session • Access to the network of the university or the internet

The usage of such amedia lab or equivalent rooms obviously leads to an enrichment of teaching and learning. Furthermore, the utilization of video streams or video conference systems giveslecturers and learners the possibility to participate remotely.However, conventional video streams normally have no real-time feedback channel. This prevents listeners of alivelecture from an active participation in terms of questions and discussions (audio and video). Video conference systems have an audio and video feedback channel buttheyare mostly complexand expensive systems. The integration of participants that are geographically spreaded is characteristic for virtual learning scenarios. Therefor,itis

73 Figure 2: Scenario and technical equipment for computer-aided face-to-face learning in amedia lab apromising approach to combine virtual environments with platforms used for face-to- face learning to achieve aﬂexible extension of the learning space on the one hand and to achieve close application to the real world for virtual learners on the other hand. The combined environments form across-technology learning community. Aprototypical system to implement across-technology learning community is presented below. The prototype interconnects amedia lab with the virtual 3D world Second Life [?]and the eLearning platform Stud.IP [dat09]. It makes possible the synchronous participation of remote listeners in the talk (passive handling) as well as in the discussion (interactive handling). Attendees may followthe live lecture on video from different an- gels of view, in case multiple audio/video services exist. At runtime, the virtual audience may choose their favorite angle of view(active handling). However, recorded lectures can just be seen in apassive and asynchronous manner.Compared to alivelecture, the video and audio sources of arecorded lecture are limited to the recorded sources.

4.2 Architectureofthe Prototype

The prototype consists of different components, shown in Figure 3together with their interconnections. Theycan be categorized and assigned to the following three layers:

• SOAconnection of the media control: The proprietary mechanisms of the media control component have to be extended by an interface for use within the service network. • Service Network: This layer represents the core of the service-based approach. Abroker as central SOAinstance is made available besides the services themselves (e.g. input and output of audio and video signals, control services). • SOAconnection of virtual learning environments: The platforms can be directly connected to the service network (with platform internal modiﬁcations) or indirectly via asurrogate.

74 Figure 3: The basic architecture of the prototype

The architecture outline demonstrates clearly that almost anyservice type, provider,or consumer can be integrated into the system easily without structural system modifications, because of the abstraction in terms of the service interfaces respective thesurrogate component. The implementation of the overall system took around ahalf year and isn’tfinished as yet. The specific solutions for every layer will be explained in the following three sections.

4.3 Integration of the Media Controller

Typically,the device services of the media lab are provided, used, and managed for face- to-face lectures without the use of aSOA.Acentral media controller allocates the speciﬁc data streams. This is the standard practice in such environments. Forinstance, the media control directs the video output stream from the laptop to the input of the projectors, and the audio output stream from the microphones to the input of the speakers. In our media lab aNI-3100 Integrated Controller [AMX06] is utilized as media controller.The controller routes media lab internal data transmissions (e.g. from camera to streaming server). The functions of this controller have been ported to the platform independent Java technology.The link between Java and the NI-3100 does not measurable limit the NI-3100s performance. The resulting Java interface has been used to implement one control service using the WebService technology [WCL05]. Furthermore, the system architecture illustrated in Figure 3includes an exemplary streaming service which preprocesses audio and video content (e.g. from camera or microphone) into streams, provided for external retrieval. In addition to live streams, the service provides access to archivedlecture recordings. The technical implementation of this differentiation is transparent for service consumers. The next section describes the SOArealization. It focuses mainly on the service provision in the media lab as well as the external consumption.

75 4.4 Propagation as aNetwork of Services

The service broker depicted above is the central component of the SOA. Even though services themselevesare also available for external use without abroker,itincreases the flexbility of service discovery dramatically.The consumers do not need apriori knowledge to address each available service, theyonly need to knowthe required service type and abroker address. In educational settings there is usually adedicated central server givenand known to the clients, so it can be used as broker.Furthermore, the service description on the broker may contain additional information that helps to decide if aservice is appropriate to an intended purpose or not (e.g. resolution of the video stream). In the presented prototype the services are registered at aWSInspection-based broker [BBM+01]. This broker is the central contact point for the external consumers that ex- plore services in the media lab.WS-Inspect savesthe services information as XML-based list that can easily be accessed via Hypertext Transfer Protocol (HTTP). Hence, an easy integration of broker requests into awide range of consumer systems is possible. The mechanisms used to invoke aservice are specific for each service. Forinstance, the streaming service uses the Real-Time Streaming Protocol (RTSP), apopular protocol to control video streams. RTSP-based video streams can, for instance, be used in Second Life and manyvideo players. The service specific invocation is an important aspect that has to be considered for the integration of consumer systems into the service network. The following section describes one consumer system, the virtual world Second Life, in detail.

4.5 Integration of aVirtual World

The 3D virtual environment Second Life is hosted by Linden Labs on aserver farm named The Grid. Currently there are over3.000 simulation servers that manage over11.000 regions (65.535 vitual qm per region), and their database servers store approximately 50 Terabyte. The physic engine HavokPhysics is appliedtosimulate realistic behavior for animations, object movesand other physical effects. The free client software presents the data receivedfrom simulation servers as a3Denvironment to the user.Furthermore, it allows users to interact with the virtual world as well as other users (acting through their avatars). In comparison to conventional telepresence scenarios [TIM+04] Second Life affords ahigher immersion and is accessible for everybody with the free client software. The added value avirtual world provides in contrast to conventional face-to-face teaching is not only to offer acopyofaclassroom or lab setting, and to broadcast alectureinthe Web. Of course, this scenario is important especially for unexperienced users –teachers as well as students –inorder to orientate oneself. Abig advantage is the almost unlim- ited changeability of the environment. From an educational point of view, this prevents the teacher from circumstantially explaining unknown situations (e.g. for learning foreign languages) or tedious theory (e.g. trafﬁc rules for adriving license) –anappropriate scenario can be simply created for the students.

76 Figure 4: Avirtual version of historic lighthouse and Teepott serves as virtual learning and teaching environment.

We decided in favour of twofamous landmarks of Rostock Warnemuende: The historic lighthouse and the Teepott as illustrated in figure 4. As the interior of the buildings should reflect the purpose of buildings and elements, and encourage an (inter)active participation, we modelled avirtual media lab with table, several chairs, and acanvastobuild an open learning and communication environment. Our virtual representation can be found in the north of the European University Island [Uni09]. The service-based communication with other environments should occur seamlessly.We use an access model based on the Second Life group concept. Registered members of our group are allowed to control the virtual and real equipment as trustable avatars. Theycan also authorize other avatars for specific events/lectures. Guest avatars can only consume content and communicate with each other or the lecturer by text and voice chat. The real media lab is controlled by asocalled Head Up Display (HUD) which appears when an avatar touches adedicated 3D object as shown in figure 5. The HUD dialog shows the different sources and drains for multimedia signals and the user can assign signal routings. This allows for the selection between different video sources (e.g. camera or slideshow) for the video stream displayed on the virtual canvas. An Linden Scripting Language (LSL) script attached to the virtual 3D object sends HTTP requests to asurrogate that bridges between Second Life and the SOA. This intermediate step is required as Second Life does not support external SOAs. An adaption to aSOA is impossible, because Second Life is not open source and the server-side software is not accessible for extensions. Messages from the real media lab are also bridged by the surrogate to allow processing within Second Life. The HTTP request contains the name of the controlling avatar,the requested service, the requested service method, and optional parameters. The receivedresponse is handled by an HTTP response handler,implemented in LSL. Possible reactions on aresponse are auser notification via text chat or changes in object appearance or behaviour. This mechanisms allowremote listeners, for instance, to select between different video sources of the media lab to be shown on the virtual canvas(e.g. video stream of the lecturer,the audience, or archivedrecordings). In the other direction, the Second Life client software shows the virtual content (e.g. audience or virtual 3D objects) to on-site

77 Figure 5: AHUD dialog appears when an avatar touches adedicated 3D object (keyboard) listeners and the lecturer.More details of the Second Life Integration can be found in [LRZ+08]. The prototypical integration of Second Life closely interweavesthe tangibility of real-life objects with elements in avirtual 3D world. Such an intuitive interaction with the learning environment exhausts the cognitive capabilities of students to amuch larger extent than in traditional classroom settings, where learners are typically acting in amuch more passive and less individual way. Several other platforms for virtual worlds exist beside Second Life. Some of them are even open source as for instance the OpenSimulator [Ope09] and Project Wonderland [Sun09]. Therefor,amediative surrogate would not be necessary.Unfortunately,these solutions are in their infancyand not as popular as Second Life. Therefor,Second Life is promising to be the best accepted platform for now. Besides the virtual world, we integrated twoother platforms into our cross-technology learning community to proof our concept: The classic eLearning platform Stud.IP as another (HTTP-based) virtual environment [GZLT08] and individual on-site cellphone clients for individual cross-environment messaging [ZDLT09].

5Summary and Further Developments

The service-based linking of face-to-face learning scenarios and different virtual learning environments described in this article goes farbeyond previous approaches that address the combination of devices and interfaces for learning. Connections between Second Life and traditional eLearning platforms already exist [KL06], buttheyare restricted to static hypertext references or acommon database. There are no systematic approaches or standards to combine synchronous and asynchronous learning paradigms for avariety of targeted platforms. Asystematic combination of real-life and virtual interaction is promising ahuge beneﬁt for electronic learning, in terms of (not only virtually) tangible E-Learning interfaces that enrich the experiences of learners –and probably also those of teachers. The presented system achievesaﬂexible and systematic coupling of platforms and tools for face-to-face and virtual teaching and learning. Consequently,itmakes use of aservice- oriented architecture (SOA). Therefor,the prototype is easy extendable concerning other platforms and tools. An intermediate service layer between the different environments

78 contains all services that are provided by the specific platforms. Thus, anyenvironment is able to consume services available in this layer.For the first time, learner and lecturer can shape the specific learning and teaching processes in an ad hoc manner beyond predefined phases (Blended Learning) or environments (decoupled face-to-face and virtual learning). This can be modified during the lecture, dynamically.Inaddition, the emerging independencyallows aunification of synchronous and asynchronous learning and teaching scenarios. This can be realized across different educational providers. An interference of administrative areas of responsibility is not longer required thanks to the transparent encapsulation of aSOA. Nevertheless, the prototype can and will be extended at several points. First of all, amore natural interaction patterns is desirable, e.g. touching media devices (screen, speaker) instead of keyboard and dialogue. Moreover, aservice-based feedback channel from the virtual learning environment into the face-to-face learning environment does not yet exist. Although currently not required (clients and browser satisfactorily perform this task) it is desirable with regards to higher flexibility.Adirect integration of SOAmechanisms into virtual worlds likeSecond Life would also increase the flexibility of the developed system just likeanextension of virtual environments to further data formats (e.g. HTML and PDF). There are additional problems with SOAs. The SOAhype produced alarge amount of different and incompatible technologies. The WebService technology is used in our prototype. Thus, alternativebut also widely distributed technologies are ignored (e.g. Jini and Bonjour). Amechanism that bridges between separated SOAislands can significantly increase the spectrum of available teaching and learning services [ZDLT08]. Furthermore, one could think of automated avatar movement and gestures in the the virtual world, generated from capturing the on-site speaker.Finally,interaction between teachers and learners can be designed more intuitionally.This is feasible by afusion of the presented approach with principles and technologies of self organization and pervasive computing [LT08]. From the non-technical view, didactical aspects need further research. The system has been tested on dedicated events. Until now, an evaluation in regular lectures is pending. It has to be clarified if such aclose contact between face-to-face and virtual learning requires owndidactical strategies for the involved types of teaching and learning. Can the traditional didactics of face-to-face learning be used or do we require novelapproaches? Howdolearners and lecturers experience the linking? This question will be answered by comprehensive practical tests with intensive pedagogical supervision in the upcoming term.

79 References

[AMX06] AMX. Hardware Reference Guide NXI-x100 Series, http://www.amx.com/techdocs/NI- X100.HardwareReferenceGuide.pdf, 2006.

[BBM+01] K. Ballinger,P.Brittenham, A. Malhotra, W. A. Nagy,and S. Pharies. WebServices Inspection Language (WS-Inspection) 1.0, GXA Speciﬁcation, http://www.serviceoriented.org/ws-inspect.html, November 2001.

[COT07] J. Cross, T. O’Driscoll, and E. Trondsen. Another Life: Virtual Worlds as Tools for Learning. eLearn Magazine,FeatureArticle,2007.

[dat09] data-quest GmbH. Stud.IP Portal Site, http://www.studip.de, 2009.

[Gol07] R. Gollup. Second life and education. Crossroads,14:4–11, December 2007.

[Gru07] Grunwald Associates LLC. Creating and Connecting: Research and Guidelines on On- line Social and Educational Networking. Technical report, National School Boards As- sociation, 2007.

[GZLT08] M. Glaeser,R.Zender,U.Lucke, and D. Tavangarian. Service-basierte Integration dy- namischer,interaktiverMedien in die Lernplattform Stud.IP. LectureNotes in Informat- ics (LNI) -Proceedings Series of the Gesellschaft fuer Informatik (GI),P-132, September 2008.

[JMLL00] A. Johnson, T. Moher,J.Leigh, and Y. Lin. QuickWorlds: Teacher drivenVRworlds in an Elementary School Curriculum. In SIGGRAPH 2000 EducatorsProgram,pages 60–63, NewOrleans, LA, USA, July 2000.

[KL06] J. Kemp and D. Livingstone. Putting aSecond Life Metaverse Skin on Learning Man- agement Systems. In Proceedings of the Second Life Education Workshop,pages 13–18, San Francisco, CA, USA, August 2006. The University of Paisley.

[Lin09a] Linden Lab. HowMeeting In Second Life Transformed IBMs Technology Elite Into Virtual World Believers. Technical report, Linden Research, Inc., 2009.

[Lin09b] Linden Lab. Second Life: Ofﬁcial site of the 3D online virtual world, http://secondlife.com, 2009.

[LRZ+08] S. Lindemann, T. Reichelt, R. Zender,U.Lucke, and D. Tavangarian. Neue E-Learning Szenarien durch bidirektionale Kopplung vonPraesenzlehre und Second Life. In Work- shop Proceedings der Tagungen MenschComputer 2008, DeLFI 2008 und Cognitive Design 2008,pages 322–326, Luebeck, Germany, September 2008. Logos Verlag Berlin.

[LT08] U. Luckeand D. Tavangarian. Mehr als Mobiles Lernen –Innovative Infrastrukturen und Dienste fur¨ Pervasive Learning. In Zeitschrift fur¨ e-Learning,volume IV/2007, pages 29–41, Wien, Austria, January 2008. Studienverlag.

[Ope09] OpenSim Team. OpenSimulator Portal Site, http://opensimulator.org, 2009.

[RH08] T. Ritzema and B. Harris. The use of Second Life for distance education. Journal of Computing Sciences in Colleges,23:110–116, June 2008.

[Sun09] Sun Microsystems. Project Wonderland Home, https://lg3d-wonderland.dev.java.net, 2009.

80 [TIM+04] N. Takamizawa,H.Ichisawa,Y.Murayama, J. Smith, and J. P. Kumquat. Use of telepresence in informal broadcasting overthe internet. In Proceedings of the 2004 ACMSIGMM workshop on Effective telepresence,pages 12–15, NewYork, NY,USA, October 2004. ACM.

[Uni09] University of Rostock. Lehren und Lernen mit virtuellen Welten, http://wwwra.informatik.uni-rostock.de/343.0.html, 2009.

[WCL05] S. Weerawarana, F. Curbera, and F. Leymann. WebServices Platform Architecture: Soap, WSDL, WS-Policy,WS-Addressing,WS-Bpel, WS-Reliable Messaging and More.Pren- tice Hall International, 2005.

[ZDLT08] R. Zender,E.Dressler,U.Lucke, and D. Tavangarian. Meta-Service Organization for a Pervasive University. In Proceedings of PerEL 2008, Workshop at 6th IEEE International Conference on Pervasive Computing and Communications (PerCom),pages 400–405, Hong Kong, China, March 2008. IEEE Computer Society.

[ZDLT09] R. Zender,E.Dressler,U.Lucke, and D. Tavangarian. Pervasive Media and Messaging Services for Immersive Learning Experiences. In Proceedings of PerEL 2009, Work- shop at 7th IEEE International Conference on Pervasive Computing and Communica- tions (PerCom),Galveston, TX, USA, March 2009. IEEE Computer Society.

Session3

Communities on the Move

The Development of aPersonal Mobile GIS

K.H. Ng, Wallace K.S. Tang

Department of ElectronicEngineering City University of Hong Kong Tat Chee Avenue, Kowloon, Hong Kong [email protected] [email protected]

Abstract: In this paper, the developmentofapersonal mobile Geographic InformationSystemisdescribed. Thesystem is configured in aclient-server architecture,involving mobile clients and aservicing server. With the use of Global Positioning System, the locations of agroup of mobile clients are identified andthese pieces of information are sent to server for additional services.For demonstration,the system is used as areal-timelocation trackingsystem.Itnot only provides thetrackingofthe users, but alsoserves as amonitoring system for issuingemergency call,whenever an abnormal scenerio is identified.These designswellillustrate howthe mobile and Internet technologies can be used to enhance this kindofapplications.The same frameworkhas also been extended for fleet management to demonstrate its potential applications in logistics industry.

1Introduction

Geographic Information System is an integration of hardwareand software for the access of spatialdataand services.Recently, with the advance development of mobile communications andInternetTechnologies, mobile GIS has beenwidelyexplored [FWZ08,LPH08]. Together with theGlobalPositon System(GPS), themobile GIS forms apowerfultoolfor real-time positioningand navigations.Thiskindofdataalso servesasakey element for the success of many location-based applications [Ab97,RM03],suchasIntelligenttransportation, on-spottourguide,emergencyservices and so on.

In this paper, it is aimedtoshare our experience in designing alow-costpersonal mobile GIS, which is simple andease to use. Thesystem can help to manage theserious accidents frequently happened on hikers and otheroutdoor sport participants in the countryside, forwhich life is lost due to thedifficulties of spotting their locations. Moreover, it helpstoidentify thelocation of elderly if necessary.

85 The proposed systemisformulated in aclient-serverarchitecture, providing services for amaximum of 10 mobile clientswith aserver. Clientscan view andkeeptrackoftheir ownand theothers’ locationsonrequest by theirmobile phones, while thesame informationisavailable at theserver side or the associated webpage.

The proposed system can be easily extendedtoother location-related applicationsand services, such as afleetmanagementfor asmall logistics companyasillustrated in this paper. It is also demonstratedhow theavailable services in Internet (such as GoogleMaps) can be used to facilitate our design.

The organizationofthispaper is as follows.InSect.2,the fundamental technologies adopted in this system are briefly described. Thedesign andthe operations of the personal GIS are then explained in details in Sect. 3. In Sect. 4, the design is furhter extendedfor theapplicationoffleet management.Finally, conclusion remarks are drawn in Sect. 5.

2Technological Background

2.1GlobalPositioning System

Nowadays, theglobal positioning system (GPS) has been widelyusedinour daily life. In order to locate thelongitudeand latitude of aterminal, at least3satellites’ data is required.Inthe case that altitude is needed,foursatellites’ data is demanded.Our system is assumedtobeoperated underanenvironment that GPS is well-utilized,and the National Marine Electronics Associationtransmission standard, NMEA0183,isadopted forreal-timepositioning.NMEA0183iscurrently themostcommonprotocol forGPS receivers in themarket. There are six different formats for NEMA data sentences,where onlytwo,namelythe GPGGA andGPRMC sentences, aretobeused in oursystem. FromGPRMC sentence, theUTC time,status, latitude,latitude direction, longitude, longitude direction, speed(in knot), headdirection anddateare obtained, while horizontal dilutionofprecision, altitude,satellite used are acquired fromGPGGA sentence.

2.2Bluetooth Communication

Since some mobile units maynot include abuilt-in GPS receiver, oursystem support the use of an external connected GPS receiverbased on the Bluetoothprotocol. Bluetoothis awirelesscommunicationstandard,operating at apublic frequency channel with frequency of 2.4GHz,providing amaximum datarateof3Mbps at amaximum distance of 10 meters between thebluetoothdevices.

By using the interface of Discovery Listener,mobile modulecan carry outsearching and querying of Bluetooth devicesnearby.Itscans allpossibleBluetoothdevices using the JSR-82standard, andbysetting“BluetoothMajor Device Classes=7936”, non-GPS receiverscan be filtered out.

86 Once adevice with class of 7936 (i.e.aGPS receiver), the mobile program inquires for theBluetoothaddress,and establishes connectioninthe read-onlymode with Stream Connection.Dataisthenread continuously using Input StreamReader.Itshould be noticed that the baud rate is not necessary to be set up,but theGPS will transmit the NMEA0183sentences in serial with atransmission rateof4800bps.

2.3 General PacketRadioService

The General Packet RadioService (GPRS) introduces packet switching onto aGSM network.Since it onlyoccupies thenetwork resource when there is adatatransmission, it is more cost effective. In this project, GPRS is usedfor thecommunicationbetweena mobile unitand theserver.

2.4ShortMessage Service

Short message service (SMS) is nowwidelyusedfor thecommunications between mobiles in textmode,asits cost is much lower than thevoice service. ASMS messageis limitedto140 alphanumericcharacters. However, it is good enoughfor using SMSto send alertmessage to thedesignatedcontact person by theserver, oncethe predefined abnormalbehaviourisidentified. Forexample,ahiker staysinaparticular location fora very long time,which probably implies theoccurrence of injuryoraccident. AGSM phoneishence constructed by AT commands to send“warming” to adesignated phone number: AT↵ AT+CMGF=1↵ AT+CMGS=“12345678” ↵warming↵ctrl+z↵

2.5GoogleMaps

GoogleMaps is afree mapservice providedbyGoogle, forwhich twokindsofimages, vector mapand satellite photos versions, areavailable.Inorder to further promote the service, Googleprovidesaset of API [Go1]thatallows users to combinethe map servicewith theirwebpage,and so non-commercials users can obtain themap as an image based on URLparameterssent throughastandardHTTP request.For mobile user, images of 320x240 pixelsare downloaded (as showninFig.3). Similarmethod is also appliedinordertodevelop thecorresponding locationtracking website forour system (See Fig. 6).

Someotherusefuland informative services arealsoavailable in GoogleMaps. For example, thedriving routebetweentwo locations andthe total distance canbeobtained: vardirections =new google.maps.Directions(); directions.load("from: 26 Jordan Road, Kowloon,Hong Kong to:83Tat Chee Avenue, Kowloon, Hong Kong"); google.maps.Event.addListener(directions,"load", function() { document.body.appendChild(document.createTextNode(directions.getDistance().meters)); Therecentdeveloped Google AJAX APIfurtherenhances theservice by allowing loading-on-demandservice [Go2].

87 3ADesign of Personal Mobile GIS

3.1SystemOverview

The overall architecture of oursystemisdepictedinFig.1.Itincludes twomainparts: clientshaving theirown mobile phones with GPS devices (external or internal)and a Desktop Server. Thefollowingsdescribe thebasicoperations.

Figure 1. The architecture of thepersonal mobile GIS

Firstly,the client programconnects to aBluetooth GPS receiver or abuilt-in GPS moduletoobtainthe real-timelocationofthe user.Based on themap calibration data and thereceivedGPS data, theprogram spots the current location of the user by plotting adot onto themap image. Therefreshment rate is per 5seconds. At the same time, user can choose whether he would liketoreport this informationtothe desktopserver viaa cellular phonedata network fortracking purpose. One canalso use it as atwo-way GPS mapping device,identifyinghis owncurrent location.

Once the severprogram received theincomingtraffic andafter data processing, the sender’scurrent location will be trackedontothe desktop screen. The serverprogram also manages any location enquiryand uploads theinformation to adesignated web server utilizingaGoogleMaps service. People can hencetrack those mobile phoneusers by simplybrowsing through internet.

Amonitoring system has also been developed.Ifthe location of aparticular userremains at the sameplace for too long (exceeding somedesignated period of time), the server will instruct its attached mobile phone to send out an alert SMS by AT commands via USBconnection.

88 3.2Client ProgramDesign

Both theclient andserver programsare developed with Javafor portability. Figure 2 depictsthe flow diagram of ourdesign,respectively. To run theclient program,amobile phonewith capabilitiesinJavaCLDC 1.1, MIDlet 2.0isneeded.The developed system hasbeentested in Motorola E2,Nokia6120c and SE K810i.

After starting theprogram,the userisrequiredtoenter hisname, theIPand port of the server andthe time interval forreportingpositiondata,asshown in Fig.3(a). Once this data inputted,itistobesaved in themobile phoneand shown automatically next time. If all thefieldsare left blank,the program will still continue,but location reporting will be disabled.

The mobile will then searchfor theBluetoothGPS receiver device.Ifthe receiver is on, thescreen will be updated as shown in Fig. 3(b). With theoption selection, user can now report its location to server depending on themobile phone system default access point,suchasGPRSor3GorWi-Fi by setting iNet On as showninFig. 3(c).Figure4 (a) depicts the case when internet connection is on (“iNet On”and thetimeoflatest transmissionwill be shown at theright-bottom corner).

Theuser’s locationistobeupdated onto themap imageevery 5secondsasshownin Fig.4(a), by continuously reading GPS data fromeither an external or abuilt-in GPS receiver, till the user turns offthe function.Atany time,the user canturnon/offboth GPS and locationreporting functions.

The locationsofthe other users registered in theservercan also be obtained by entering into Query mode.Figure 4(b) shows thelist obtained by querying theserver about all theother users, and thelocationofanother user (say May) is then shown as Fig. 4(c) by pressing thecorresponding number.

3.3 ServerProgram Design

Theflow diagramofthe serverprogram is depicted in Fig. 5. It reads the user configuration file during start-up.Ifthe configuration file is missingorinvalid, the program will prompt theuser to input thedataagain. Thisinformation is importantfor the tracking purpose, anddue to thelimitationofpages, only theusers need “GPS mapping”are considered.

Additionalutility functionsare also developed in thesystem to allowuserfor calibrating theirmap images and generating themobile jar-program automatically.

89 Figure 2. Flow diagram for the mobile unit

90 Figure 3. (a)Loginpage (b)Startingpage(c) Optionselectionfor INet On

Figure 4. Locationtracking on mobile phone(a) user’s location(b) query of otherusers (c) location of another user

The users’locations arereceived by theserver whenthe Internetlistening function is ‘On’ (the useralsoneeds to turn on thetracking functionathis mobileside). The processed positioninformationwill then be displayedbothgraphically on themap andin text. Theold reported locationsare in blue,while thelatest one is in yellow, as depicted in Fig. 6(a).The trace of aparticular usercan be loggedinagpx format, which is XML basedformatfor positiondataexchange. Thesystemsupportsamaximum number of 10 users.The server also manages any location query from auser by replyingall theother users’ locationtothe requester.

In particular to safeguard thehiker,aspecial emergency SMSfunction is implemented. Forexample, if themobility node stays in an area of 100 meter radiusoverapreset period,says 30 minutes, theserver will sendout aSMS to adesignated mobile phone number via the attached GSM phone/modem as an alert.

If aweb serverisavailable, thesamelocationinformationcan be uploaded to theweb servervia FTPprotocol.Bythe use of GoogleMapsAPI [Go1], thecurrent locationsof theusers can be monitoredonGoogleMaps viathe html page from theweb server, as shown in Fig. 6(b).

91 Figure5.Flow diagram for the server program

Figure 6. Location of users displayed (a) at serverside(b) onto GoogleMaps

92 4Extended Applications

The application of ourframework canbeeasily extendedfor theuse in other location- based services. In this section,itisappliedfor acar fleet managementsystem, well suited for asmall company.

4.1Overview

Ourdesign is modified foralow-cost car fleet managementsystem, including three major parts: the server, thedesktop clients andmobile clients.Database is implemented in theserver, in ordertodeliver jobrecords,drivers’current positions andsoon. The serve also used as abusiness logic/GIS server. Desktop client allows operationalstaff to managethe jobrecords and generate business reports. Mobile clientislocated at driver’s side, reportingcurrentdriver’s positiontoserver, acquiringnew jobfromthe server and reporting thejob statustothe server.

Database

Desktop

internet Business Logic/ GIS server

Mobile Clients withGPS receiver Figure 7. Thearchitecture of themobile GISfor fleet management

4.2Flowofthe JobManagement

Theoperational staff submits anew job to thesystem with aspecified service time. Jobsare classified into twomaintypes: short-termjob (tobeservedimmediately or hourslater)and long-term job (tobeserveddemanded in nextday(s)).

93 4.2.1The Short-TermJobs

The process on short-termjobsisdescribed as follows:

1. Server submits thepick up pointquery to Google for theretrieval of its latitude / longitude.Anexample of URL enquiry is: http://maps.google.com/maps/geo?q=83 Tat Chee Avenue, Kowloon, Hong Kong&output=csv [Go2]. 2. Theresult is then compared with thepositions of free-status drivers updatedbythe mobile clients, and their distances arecalculated. 3. Thenearestdriver will automatically be selected but this choice needs to be confirmed by theoperational staff (see Fig. 8). 4. The status of theselected driverstatus is then updated in thedatabase. 5. The Mobile Client receivesthe newjob viaregular update (see Fig. 9). 6. The drivercan confirm or reject the neworder. 7. If the driveraccepts the job, the status of the job will be updated in the database. Otherwise, thesystemwill suggest another driver to theoperational staff. On the otherhand, theoperational staffcan also assign driver manually.

Figure 8. Interface in Desktop Client

4.2.2 The Long-termJobs

Forlong-term jobs,amultiobjectiveGenetic Algorithm(GA) [YT09]can alsobe incorporated. Based on the distances between pick-up points of each job and thedepot (it is also obtained by performing thequery to GoogleMapsasdiscussed in Sect. 2.5), optimal routing of multiple vehicles can be obtained with theminimization of thetotal distance andservicing time.Figure10depicts oneofthe optimal solutions to handle 14 jobs,assumingthat there are4vehicles in adepot.

94 Figure 9. User Interface at Mobile Client: (a) Login page (b) Menu page (b) Job received at Mobile Client

Figure 10.The routes of thefourvehiclesobtained with GoogleMapsAPI

95 5Conclusion

In this paper, twogeographic informationsystems enhancedbymobile communications and Internet Technology have beenpresented.The first oneisapersonal GIS, which serves as areal-time locationtracking system.Thiscan help to monitorthe activities of hikers or outdoor activity participators,and also identify thelocation of elderly if necessary.The locationofaparticular mobile client is obtained by GPSand reportedto theserver viathe GPRS or 3G. Anyabnormal case will trigger an alarm messagetoa designated personusing SMS. Thesystem hasbeentested in Motorola E2,Nokia 6120c and SE K810i.The sameconcept can easily be extended forthe design of afleet managementsystem. Based on theGPS informationofeachdriver, an incomingjob can be suitablyand automatically assigned to thenearest driver. Moreover, aGenetic Algorithm can also be included foridentifyingthe optimal routing of vehicles for multiple jobs.Inbothcases, it is also demonstrated thatthe recentdevelopmentsin Internet service, in particular theGoogleMaps and its services, can greatly enhance these kindsofdesigns, making alow-costpersonal GIS possible.

6Acknowledgement

This work described in this paperwas fullysupported by agrant from theResearch Grants Council of theHongKongAdministrativeRegion,China (CityU 120407).

References

[Ab97] Abowd, G.D.; Atkeson, C.G.;Hong,J.; Long, S,; Kooper, R.;Pinkerton,M.: Cyberguide: Amobile context-aware tour guide,WirelessNetworks, 3(5), 421-433, Oct1997. [FWZ08] Fu, X.L.; Wang,D.L.;Zhang,M.M.: Thesolution of mobile GIS based on bluetooth GPSreceiver,4th Int.Conf. Wireless Communications, Networking and Mobile Computing, 2008. [Go1] GoogleMaps: http://code.google.com/apis/maps/documentation/introduction.html [Go2]GoogleMaps: http://code.google.com/apis/maps/documentation/geocoding/ [LPH08] Lee, H.H.; Park, I.K.;Hong, K.S.:Designand implementation of amobile devices- based real-tim location tracking,2nd Int.Conf.Mobile Ubiquitous Computing, Systems, Services and Technologies, 178-183,2008. [RM03] Rao, B.;Minakakis,L.: Eolutionofmobile location-based services, Communicaitionsofthe ACM, 46(12) 61-65, Dec 2003. [YT09] Yin, J.J.; Tang,W.K.S.: Agenetic approach for two-dimensionalloading capacitated vehicle routing problems, Dynamics of Continuous, Discrete and Impulsive Systems, Series B, accepted for publication, 2009.

96 FacebookAgent –anAgent-Enhanced Social(Mobile) Network Application

Volkmar Schau, Christian Erfurth, ReneP´ asold, Wilhelm Rossak Friedrich Schiller University of Jena, Institute of Computer Science Ernst-Abbe-Platz 2, 07743 Jena, Germany {volkmar.schau|christian.erfurth|rene.pasold|wilhelm.rossak}@uni-jena.de

Abstract: Over the past years, anew kind of understanding and utilization of Internet services has emerged. People successfully form online communities and contribute actively.Facebook is one example for such an online social network (OSN) which has attracted millions of Internet users within avery short time period. In this paper we followthe idea to expand the virtual world spanned by OSNs into the user’sreal world. To achieve this goal, we apply the mobile agent paradigm as aconnector concept to strengthen the integration of users. As an example for OSNs we rely on Facebook.

1Introduction

As we have movedfurther into the virtual world that started with the World Wide Web in its initial version we are nowacouple of years into the Web2.0 era. In the seamless transition from ”Web 1.0” to Web2.0 communication and organization have changed drastically.Today,the Webisthe most popular communication channel for accessing and sharing information. At the beginning, and to some degree even today,the Webwas highly unorganized. The start-up period of Web1.0 wascharacterized more by data assembling than information organization, which followed no general standard. We nowsee amore standardized Webenhanced by semantic meta data and network services. Using network services, the entire media spectrum is becoming available in astandardized fashion and replaces the browsing of plain Webpages. Network services open away to retrieve and combine data itemsinto newmedia and information aggregates. Online social networks (OSN) likeFacebook [Fac] and MySpace [Incb] are readytoannounce network services as newmedia. Thus, we see that there is ashift from what could be called vertical to horizontal communication. Even users participating in online social networks are empowered to announce their ownservices. Moreover, everyone nowhas avoice and the possibility to spread news, rumors, photos and links at tremendous speed easily into the public community sphere. In this article we used Facebook as the basis for anew kind of OSN extension. The online social network Facebook enables its users to present themselves in an online proﬁle, ac- cumulate friends and vieweach other’sproﬁles. Facebook members can also join groups based on common interests and create their ownenvironment by adding Facebook extensions and sharing them with their friends. The idea of the FacebookAgent application is

97 based on the results of the joint research project called MobiSoft that wasrun by Friedrich Schiller University Jena (FSU), the agent factory GmbH, and Godyo AG.The project was funded by the Thuringian Ministry of Economy,Technology and Labor.InMobiSoft we aimed at different application scenarios of personal assistants (on mobile devices) that range from information retrievaland control of legacy systems in industry to the support of human interactions in all places where people come together [EKR+08]. Especially those scenarios focused on the support of human interactions and communities attracted further research interests. With MobiSoft and our industrial strength research in the ﬁeld of mobile agents (Tracy), atechnological infrastructure for mobile agent based community applications wasestablished: The Tracyagent system executable on Java-enabled (mobile) devices, with aset of possible add-on features, and Tracyagents which are able to make multiple hops between Tracy-enabled networked devices. The integration of mobile devices into community systems is challenging. Typically the usage of amobile device happens at the spur of the moment. Social-mobile applications must, therefore, be simple to use. Mobile agents are well suited to provide essential support in this matter,due to their autonomous nature. The rest of the paper is organized as follows. Within the next section we discuss the evolution of online social networks and analyze assistant integration. In section 3we highlight FacebookAgent. In section 4wereviewthe approach and present our related results. Finally we give aconclusion.

2Related Work

Online social networks have beenthe global consumer phenomenon of 2008. Nielsen On- line points out, that social network and blogging sites are nowthe fourth most popular activity on the internet ahead of personal emails [Com09]. The story is consistent across the world and dates back to the early days of computer networks. Following the SixDe- grees.com experience in 1997, the doyen of social network sites, that allowed its users to create profiles, invite friends, organize groups, and surf other user profiles, hundreds of social networks spurred online [AG06, Gob09]. The growth of popularity is motivated in an Internet community where people spend their time interacting with others. Likemicro blogs people generate news feeds about their daily livesthat cultivate friendships. In our daily life mobile communications has amost important value. Thereby,mobile communication devicesand applications are primarily designed to increase efficiencyand productivity,aswell as to manage our fast wayoflife. However, that growth in popularity of mobile communications is only half the story.For manypeople, particularly younger users, cell phones, smart phones, and other handhelds primarily have asocial function [EP05]. Afew small companies are beginning to exploit the growing demand for social- mobile applications and start anew market of mobile social-software services (MSS). One of the first definition of social-mobile application wasgiven by Lugano (TeliaSonera Finland) as mobile-social software (MoSoSo), defined as aclass of mobile application whose scope is to support social interaction among interconnected individuals [GL:07].

98 We present here only abrief survey of the currently most popular social software services, which is only aselection of the upcoming boom. Forour approach the challenge is the aspect of facilitating, augmenting, and promoting human social interaction by electronic personal agents in the daily mobile life. One of the most popular MSSs applications is dodgeball [Inca], aNew York-based social- mobile network with thousands of users across the US. After aregistered user enters the network friends receive atextmessage indicating the check-in location and time in case theywant to get together.The service will also notify auser if afriend, or friend of afriend is checking in. In the UK, rummble [Rum] helps mobile users locate nearby friends, friends of mutual acquaintances, or even strangers with matching preferences. Yet another is Plazes [gG] alocation-aware interaction system that helps mobile users hook up with friends or other like-minded people anywhere on the globe. With match2blue [mUI] there is asocial-software service under development basedonly on mobile phones. The service allows to access profile matchers wherevermobile social network users happen to be. Users can makecontact with other mobile phone users who share similar interests. Jambo Networks [Net] focuses on widely-used mobile devices likeWi-Fi-enabled laptops, cell phones, and PDAs to match people within walking distance who have similar interests and would liketomeet face to face. However, in anutshell we could not find anyfurther support by personal assistants. Additionally,overthe past years, Intel Research Seattle has designed, studied, and built several social applications likeHouston [CESL06, oWS, Smi05], an application designed to investigate the utility of mobile social-support networks. Houston is oriented towards physical fitness and weight management, butthe general principles apply to manyother areas where friends share experiences and generate mutual support. With Houston, group members share step counts from their pedometers automatically via mobile phones for example, ”Joe made it to 10,000 steps today!”. We are already sharing our daily life experiences with the community,aswith twitter,where prominent events like”There’s aplane in the Hudson. I’m on the ferry going to pick up the people. Crazy.” are posted. [Kru09]. Givenbyaselection of most popular social services, we are in the starting blocks to use online social networks as living social communities. The number of social networks is growing fast. These sites account for one in every 11 minutes going online [Com09] and people spend more and more time online, managing by hand their social communities. In [EKR+08, KBR06] Erfurth and Kern present as an alternative social mobile assistants as waytomanage MSSs. Porkahr et al. endorse this path by proposing ”Agents: Technol- ogy for the Mainstream?” [PBL05]. Acombination of these twofields –OSNs and mobile agents –could not be found in literature and in OSNs we analyzed. There are no facilitating, augmenting, and promoting human social interactions assisted by personal agents. However, we expect efforts there within the next years.

99 3FacebookAgent

FacebookAgent is asocial-mobile application for the OSN Facebook. With Facebook- Agent aFacebook member is linked with Facebook via apersonalized mobile software agent. This agent is avirtualsubstitute for the member and is able to update the member’s personal data (e. g. current location) autonomously and inform the user on events arising at the Facebook site. FacebookAgent is also runnable on mobile devices to support mobile community members. It uses Facebook features and carries them to the user’sdevice. Ad- ditional community features outside of Facebook are also possible (e. g. finding interesting people nearby) butnot yet implemented. Forour agent-based community experiment we choose Facebook as an OSN due to the fact that it is very popular,especially with younger people. Facebook, created in 2004, reports to have more than 175 million active users. The site is used on avery regular base by its users: The typical user spends about 20 minutes aday on the site, and two-thirds of users log in at least once aday [Cas06]. After its success among college students, Facebook launched ahigh school version in 2005. In the following year,the company introduced communities for commercial organizations, afeature that wasused by 22,000 organizations within the same year [Smi06]. Facebook itself has anumber of community features. The Wall is aspace on every user’s profile page that allows friends to post messages and attachments for the user.The Pokes feature allows users to send avirtual ”poke” to each other (a notification that tells auser that theyhavebeen poked). Photos and albums can be uploaded with the Photos feature. There is also a Status,which informs user’sfriends of their whereabouts and actions.The Wall of auser is visible to other Facebook members who are able to see that user’sprofile (depends also on user’sprivacy settings). News Feed on every user’shomepage shows configurable information likeupcoming events and birthdays related to the user’sfriends. The mentioned Facebook features are, or rather can be, utilized by the FacebookAgent application. The content is controlled or modified by agents on behalf of the user.

3.1 Applications integration on the Facebook platform

Facebook offers asophisticated plug-in system, which givesthe opportunity to integrate your ownapplications into the platform. This wayitispossible to integrate other applications, platforms or services with Facebook and –ifthe installing user permits this –access the user’sdata andother services from those. Applications can be found using the search function in Facebook. Alternatively the user can choose from the product directory.Ad- ditionally one might recommend an application to friends, when installing, though most application adds seem to stem from the proﬁle boxes [Smi07]. Facebook’sREST-likeinterface can be used by multiple client libraries or APIs, that are usually provided as open source. Ofﬁcial libraries are offered for PHP and JavaScript, butthere are various third party libraries available for Java,C#, Ruby and manymore [Fac08a]. In order to connect to the Facebook Server every application has to be registered

100 101 102 Tracy

Layer Agent Agent Agent Agent nt Age ... Shell Plugin Layer Migration Authorisation Social Network Communication

Kernel

JavaVM

Figure 3: TracyArchitecture

3.3 Agent System Toolkit

Over the past years, we’ve developed the Agent System Toolkits Tracyasversion named Tracy2SE and Tracy2ME. The latter one naturally benefited from our experiences with the Tracy2SE version being alightweight version of the former for mobile devices. Tracy2SE relies on amicro kernel architecture, compare Figure 3, whereas its predecessor applied a strict 3-tier design which provedtobefar to monolithic in action. On top of the kernel, anumber of plug-ins provide the actual functionality of aTracysystem. Forexample, message exchange among agents, agent migration, or several security mechanisms are realized as plug-ins. As with all plug-ins, the kernel is responsible to control the agent life cycle and to coordinate interactions among agents and plug-ins. Over the last years, this highly modular and extensible architecture provedtobeuseful, as it is very easy to adapt abase Tracysystem to custom needs by simply defining the set of necessary plug- ins and corresponding agents. One of the main aims of the mentioned MobiSoft Project [EKR+08] wastoport Tracy2 to the Java Micro Edition environment to allowfor the usage of agents and agent migration on mobile devices. MobiSoft targeted to host agents on mobile devicesand has enabled agent migration between mobile phones via Bluetooth and to distant hosts overGPRS, UMTS or Wi-Fi. Looking at the architecture, Tracyisdivided into three major components: the agent layer, the plug-in layer and the kernel as their foundation. Plug-ins and agents followspecific design rules described in [BR05] and [Sch08]. Both Tracy2SE and Tracy2ME provide at least one agency, which is used by different kinds of agents. In our FacebookAgent application, there are twodifferent agent-types: Multiple mobile Tracy2ME agents, that im-

103 Facebook

App Profiles HTTP

JSON REST-Server

JSON

Tracy Tomcat API API :FP :SUPP SOAP FAW

AM :SNP SOAP

new services

... Data

Figure 4: Communication system

plement one function per agent and the corresponding GUI,and one stationary Tracy2SE agent, that is able to communicate with the ME-agents and use the SocialNetwork plug-in (see Sec. 3.4) to communicate with Facebook.

3.4 Plug-in System and Communication

Forthe FacebookAgent application three different Tracyplug-ins have been developed to couple Facebook with our agents. The main plug-in is SocialNetwork (SNP), which is designed as ageneric social networking solution. It makes use of the platform-specific Facebook plug-in (FP) that uses the facebook-java-api (see [Fac08b]) to access Facebook data and functions. The third plug-in SocialUserProfiles (SUPP) is used to store information about the user,his connections and the corresponding agent in auser profile. As Facebook requires user interaction during the login process even when using the API, the Facebook plug-in acts likeascripted browser to circumvent the manual, that is browser drivenlogin process. That way, agents can login in afashion that is transparent for the user. The Facebook application Facebook AgentWall (FAW)can be used to configure the agents behaviour and represents the system towards other Facebook users. It’sthe starting point of the installation process and communicates with the SocialUserProfiles and SocialNet- work plug-ins through Webservices to store and retrieve configuration data. Additionally

104 it spawns anew Tracy2SE agent during the installation, which is needed for the communication with the mobile Tracy2ME agents. It’simplemented as aJavaservlet and uses FBML-containing Java server pages as templates.

4Assessment of FacebookAgent and Related Endeavors

With the exemplary combination of Facebook and mobile agents we target our research on aseamless integration of human life and OSNs with aspecial focus on mobile users. With other words, mobile agents can be used to expand the virtual world spanned by OSNs into the user’sreal world. Humans which liketobepart of an online community do not need to be online in person. Personalized agents care for their integration. Thinking further, persons who get in contact ofﬂine can be connected online too by exchanging OSN contact details, using mobile agents. Agents can also be used to manage personal information provided in OSNs. In the case where humans belong to different communities, agents may deploythis information and may care for consistent data on the OSNs. In another student project we have developed acombination of Second Life and mobile agents. In such avirtual world characters monitoredorcontrolled by agents can react to inquiries or forward these to humans. Agents are even able to takeoverthe execution of simple tasks within the virtual world or/and within the networked OSNs. Forfuture mobile community applications more non-technical aspects in the social area of potential users are also relevant e. g.

• Has usability reached an appropriate levelfor ad hoc mobile device usage? • Is there acorrelation between age and gender of potential users and the acceptance of mobile application usage? • Is the usage of mobile devices as acommunity interface accepted?

Consolidated ﬁndings in this interdisciplinary area are essential for asuccessful application of future technologies in the mobile sector.The user is the central element in ubiquitous computing. As part of the MobiSoft project, we made asurvey on campus to investigate the acceptance of personal electronic assistants. We receivedover1000 sub- missions of the online questionary (70% from students, 23% from staff). We asked the participants to rank the importance of possibly available newservices when using software assistants. For70% of the participants services with software assistants in the library sector (one application scenario in the project) were rated as important or very important. In contrast to that, 61% of participants stated that it would be less important or not important at all to have assistants for cafeteria menu information on mobile devices (another sample scenario). Social-mobile assistants are generally ranked as at least partially important. The readi- ness of use is ranked nearly identical for all application areas. Despite of the fact that the newtechnology of mobile personal assistants is not yet well known, about 50% of the

105 45% 40% 35% 30% 25% 20% 15% 10% 5% very important 0% important partly important little important unimportant

Figure 5: Survey Results: Importance of Selected Mobile Applications

interviewed persons answered that theywould perhaps use assistants. Fig. 5presents the results of the question to rank the importance of apossible application of mobile software assistants in selected areas typically found at auniversity.Apparently the scenario ”learning groups” is astrong community application. About 1/3 of interviewees answered that this scenario would be very important or at least important. Another third of the population sees this scenario as partly important. The aim and the systematic approach of MobiSoft, whereupon this work is based, we introduce in [EKR+08]. The results of the questinary helped and motivated us to investigate the combination of mobile agents and communities in more detail. Forsure the questionary is no evaluation for our experiment. However, we did not go into an evaluation of the FacebookAgent because it is hard, or even unclear,how an evaluation could look like. Forsuch acommunity application, only the community can evaluate newfeatures and accept or discard them. This is similar to applications in the Web2.0 area which are often released as beta versions and improvedifworthwhile.

5Conclusion

FacebookAgent is one scenario to investigate the application of mobile agents in the context of community systems. As mentioned above,the MobiSoft project has already established asuitable technological basis in this context and has delivered aﬁrst experience in

106 the community field. The experimentalprototypes we used focused on the usage of agents on mobile devices to support mobile and distributed communities. With FacebookAgent an example for the combination of mobile agents and OSNs is realized. Taking this combination and the results of the MobiSoft project, the usage of OSNs by mobile community members can reach anew level: With personalized mobile assistants the integration of OSN community members in every day life is strengthened. Up-to-date information can be provided within OSNs by electronic assistants autonomously and –vice versa –users will be well-informed by their assistants. Since 1998 we have been continuing our strategy in autonomous operating agents. Initial start-up givenbyProf. Rossak we achievedour first goal in 2000 and 2003 by releasing our Agent System Toolkits Tracy. Designed as aflexible system drivenbyservices new functions are integrated by service plugins[BER00, BMG+04a, BMG+04b]. Along with MobiSoft we transformed Tracyinto alightweigh Agent Toolkit running on mobile devices. Henceforth agents are able to operate into the mobile life [EES07]. By nowweare on the cusp of interdisciplinary agent work learning about social workflowcompositions (likeMSSs, OSNs and domain specific knowledge) and experiencing with the idea of life (genetic algorithm) to come closer for autonomous operating agents.

References

[AG06] Alessandro Acquisti and Ralph Gross. Imagined communities: Awareness, information sharing, and privacy on the Facebook. In In 6th Workshop on Privacy Enhancing Technologies,pages 36–58, 2006.

[BER00] Peter Braun, Christian Erfurth, and Wilhelm Rossak. An Introduction to the Tracy Mobile Agent System. Technical Report No. 2000/24, Jenaer Schriften zur Mathe- matik und Informatik, Friedrich Schiller University Jena,2000.

[BMG+04a] Peter Braun, Ingo Muller¨ ,SvenGeisenhainer,Volkmar Schau, and Wilhelm Rossak. Agent Migration as an Optional Service in an Extendable Agent Toolkit Architecture. In Ahmed Karmouch, Larry Korba, and Edmundo Roberto Mauro Madeira, editors, MATA,volume 3284 of LectureNotes in Computer Science,pages 127–136. Springer, 2004.

[BMG+04b] Peter Braun, Ingo Muller¨ ,SvenGeisenhainer,Volkmar Schau, and Wilhelm Rossak. AService-oriented Software Architecture for Mobile Agent Toolkits. In ECBS,pages 550–556. IEEE Computer Society,2004.

[BR05] Peter Braun and Wilhelm Rossak. Mobile Agents -Basic concepts, mobility models & the Tracy toolkit.dpunkt.verlag, 2005.

[Cas06] John Cassidy.The Online Life -MeMedia. The NewYorker,May 15 2006.

[CESL06] SunnyConsolvo, Katherine Everitt, Ian Smith, and James A. Landay,editors. Design requirements for technologies that encourage physical activity.ACM Press, 2006. pages 457-466.

[Com09] Nielsen Company. Global Faces and Networked Places. Nielsen Online, March 2009.

107 [EES07] Gerald Eichler,Christian Erfurth, and Volkmar Schau. Enhancing Communities by Social Interactions in Mobile Environments. July 2007. [EKR+08] Christian Erfurth, Steffen Kern, Wilhelm Rossak, Peter Braun, and Antje Leßmann. MobiSoft: Networked Personal Assistants for Mobile Users in Everyday Life. In Matthias Klusch, Michal Pechoucek, and Axel Polleres, editors, CIA,volume 5180 of LectureNotes in Computer Science,pages 147–161. Springer,2008. [EP05] N. Eagle and A. Pentland. Social serendipity: mobilizing social software. Pervasive Computing,IEEE,4(2):28–34, 2005. [Fac] Facebook. Online Social Network Facebook. http://www.facebook.com. [Fac08a] Facebook. Facebook Developers Resource. http://developers.facebook.com/resources.php, April 2008. [Fac08b] Facebook. facebook-java-api. http://code.google.com/p/facebook-java-api/, Novem- ber 2008. [gG] Nokia gate5 GmbH. Online Social WebNetwork. http://www.plazes.com. [GL:07] Mobile Social Software: Deﬁnition, Scope and Applications,Proceedings eChallenges 2007 conference. eChallenges 2007 conference, 2007. Lugano, G. [Gob09] Gord Goble. The History of Social Networking, Januar 21 2009. [Inca] Google Inc. Online Social Network Dodgeball. http://www.dodgeball.com. [Incb] MySpace Inc. Private Online Community MySpace. http://www.myspace.com. [KBR06] Steffen Kern, Peter Braun, and Wilhelm Rossak. MobiSoft: An Agent-Based Mid- dleware for Social-Mobile Applications. In Robert Meersman, Zahir Tari, and Pilar Herrero, editors, OTMWorkshops (1),volume 4277 of LectureNotes in Computer Science,pages 984–993. Springer,2006. [Kru09] Janis Krums. There’saplane in the Hudson., January,152009. http://twitpic.com/135xa. [mUI] match2blue US. Inc. Mobile Social Network match2blue. http://www.match2blue.com. [Net] Jambo Networks. Mobile Membership Directory.http://www.jambo.net. [oWS] University of Washington and Intel Research Seattle. UbiFit. http://dub.washington.edu/projects/ubiﬁt/. [PBL05] Alexander Pokahr,Lars Braubach, and Winfried Lamersdorf. Agenten: Technologie fur¨ den Mainstream? In it -Information Technology,pages 300–307. Oldenbourg Verlag, 11 2005. [Rum] Rummble. Personal Mobile Network. http://www.rummble.com. [Sch08] Volkmar Schau. How to write aTracy2 Plugin.Friedrich Schiller University of Jena, February 2008. [Smi05] Ian Smith. Social-Mobile Applications. Computer,38(4):84–85, 2005. [Smi06] J. Smith. Updated lists of all companies and regions on Facebook, November 15 2006. [Smi07] J. Smith. Newdata on Facebook application virality,October 2007.

108 Integrated Solutionsand Services in Public Transporton Mobile Devices

Karl-Heinz Lüke1,HolgerMügge2,MatthiasEisemann3,AnkeTelschow4

1,3DeutscheTelekom Laboratories, Berlin,Germany [email protected], [email protected]

2InstituteofComputerScience III, University of Bonn, Germany [email protected]

4T-SystemsEnterpriseServicesGmbH, Darmstadt,Germany [email protected]

Abstract: Oursociety is characterisedbyindividuality,comfort andmobility.It hasbeen showninmanyscientific studies that themobile phone playsan important role in ourlivingand workingenvironment. While navigationsystems in carsoffer ahighlevel of individuality,comfort andahigh degree of integration with thecar electronics,there arenocomparablesolutions andservices availablein public transport. In this paper, it is describedthatintegratedsolutions in public transportcan improvethe user needsinterms of flexibility andconvenience. Althoughthere areseveral individualmobile applications forrailinformationand ticketingavailable, an integrated andprofile-based solution is hard to find on the market. We proposeanintegrativearchitecturethatcoversmobile trip planning, intelligentmobile ticketingand community solutions during thetrip. This shows that ourfindingscan enhanceflexibility andcomfort in public transport.

1Motivation

In many situations,passengersfeelcomfortable when usingpublic transport, butonthe otherhand, they feel inflexible andmostlynot well informed aboutthe actualtrip informationatchangingpoints. It can also be detected that many customersare often annoyedwhenusing theexistingticketingsolutions,e.g.waitingatticketing machines or queueatacounter. In most cases they arenot so familiar with thecomplex handling of existing traditionaland mobile ticketingsolutions.Apart from themultifunctional mobile ticketingproject Ring&Ride, whichisintroduced in thenextsections,there are hardly anycomparabletechnical solutions combiningintermodal1 mobile ticketing both forlongand short-distance travel.2 Intermodalmobile ticketing solutions can use locationinformation, e.g.GSM,W-LAN,GPS or NFC, of mobile phones for reconstructing theroute takenand calculating thecorresponding fare.Duringthe trip,it is auniqueopportunity in public transporttoprovide passengers with informationand entertainment/community offers,e.g.music,video,communities, travel guides.

1 Usingshort andlongdistancepublictransport on atrip, forexample. 2 Thetouch&travelproject uses near-fieldcommunication (NFC)[Ba09].

109 Arecentstudy in Germany [Zu05] found that thepassengers wouldliketohavebetter supportconcerning howtoget to thestationand when to change vehicles in public transport. Otherstudies [Vi08] pointed outthatthe market of mobile social communities will grow steadilyuntil 2012.Almost25% of mobile subscriberswill usemobile social communitiesin2012. Thepotentialfor advertisinginmobile communitieswill also offer ahugeopportunity forthe operators in thenextfew years[Vi08].Duringthe trip,the passenger is responsive to “kill-time” offers.Typically,location-basedservices [Sa07; BI08]can supportthe passengerswiththe relevant informationservices they need in this situation. Beside thehugerelevance of location-basedinformation,personalisationof services [Fr06] also playsanimportant role in themobile environment. Acustomer’s information, e.g.saved travel plans, home address, costsaspectsand sightseeing, can be used in an expedientmannerfor intelligentand integrated services.

Regardingthe mobile device market,the prerequisitesfor an integrated solutionare given. Most applications areimplemented as abrowser-based or aclient-based solution on amobile operatingsystem. Themarketfor smartphones3 is growing, especially from 2007 to 2008, thenumberhas risenby28% [Ca09].

Thepaper is structured as follows:insection2,wedescribethe trip planning scenario. Thelocation-basedmobile ticketing project Ring&Rideisintroduced in section3. Section4illustrates differentaspectsofinfotainment/community in public transportthat demand an openarchitecturecapable to integratethird partyservices.Suchan architectureissketchedinsection5.Finally,section6summarisesthispaper andgives an outlook to future work.

2TripPlanning andTripManagement

In this section, theway atripcan be plannedconcerning departureand arrivaldetails will be shown.

Technological developments andincreasinglypowerfulmobile enddevices with features likeNFC or GPScan be linked with informationsystems (e.g.traffic andtourist). This opens up completely newtravelplanningpossibilitiesand provides supportduringan intermodaltrip. Theexpansion of mobile broadband networks allows forthe provisionof informationonmapsbothwithstatic anddynamiccontent at ahighbit rate.The combinationofthese technologies andtheir integrationintoexistingplatforms creates increasinglyintelligent services andsupports thecustomer’ssatisfaction.

By providingtraveldetails andmakingthe positionofthe travellerknown through locationtechnologies,e.g.GPS or W-LAN, as well as thecurrent time of departureof public transport4 andnecessary informationsuchasmapsand details aboutthe trip can be shown.

3 Forpreviousyears see[Ca07]. 4 Dynamictraveldataofmosttransportationcompanies is availablebut notintegratedinmostcases.

110 Making possibledelaysalong theway,e.g.traffic jams or detours,5 known,the traveller gets areminderwhenhehas to starthis trip.While travelling to thestation, thecurrent travel plan is continuously compared with thedynamictraffic data andpositionofthe user,and informationaboutthe current status is provided.6 At thestation, theuseris informed aboutavailableparking spotsand howpublic transportcan be reached.

When getting on public transport, forexample atrain,the context-specificinformation such as navigationwithinthe train, to find theseatreserved, therestaurant, thetrain crew or providingthe restaurant’s menu andmuchmore, is adapted. In caseofdelaysor connections,the travellerisinformedinatimelymanner, andanother routemay even be calculatedand recommended.Iftripreplanning is needed,alternativeroutesare determined andthe relevant information, e.g. dynamictraffic data or maps in stations, areprovidedtothe passenger’s device as well.7

At changing points, thetravellerisguidedalong theway by in-house localisationand mapnavigationtothe connectingpublic transport. Depending on thelengthofhis stopover,further information, e.g.onmuseums or restaurants, canbeprovidedaswell. Thetravelleristhenaccompaniedfromthe finalstationtothe destinationoriginally entered.

Ahighlevel of usabilityfacilitates theacceptanceofaservice. Attentionshouldalsobe paid to an easyinstallationonthe customer´s device. An exampleofinteractivehandling is Apple’sAppstore fordownloading andinstalling as well as usingthe iPhone services. Theusermustnot feel overwhelmedorevenbothereddue to aflood of information. Anotherbenefit of adequate user interactionisthe possibilityofreceiving relevant informationeithersentbythe operator(push mode)orrequested activelybythe traveller (pullmode), achoicemadebythe user himself, dependingonthe situation, forexample if making aconnectionbecomescritical.Further important aspectsare theuserinterface andinteractiondesignfor easynavigationthrough theservice andclear presentationof theinterface on arestricteddisplay size of existing mobile devices.Byintegrating personal data,the trip can be tailoredspecifically to thetraveller’s needs. Preferences likereserving awindowseat, thequickestormostconvenientconnections or an interestingentertainment programcan be takenintoconsideration.

Currently,there areanumberofapplications fortripplanning, e.g.Fahrplan(iPhone), ZugInfo(iPhone), FahrInfo VBB(iPhone), SBBFahrplan(iPhone),8 DB Railnavigator,9 availableonthe market.Someofthese applications includelocationfeatures for determiningthe positionofthe traveler,but dynamictraffic data is currently integrated in an inadequate way.

5 http://www.adac.de/Verkehr/mobiledienste/default.asporhttp://mobil.verkehrsinfo.de/ 6 Aproject at T-Labs wasconductedbythe University of Bonn whichdealt with time management while travelingtoastation. Thestatuswas shownasagreen,yelloworred light, forexample,depicting more than enough timetoreach thestationontime, beingshort of time, andnot having enough timeatall, respectively. 7 Fortripreplanningsee [MLE07]. 8 Forapplications forthe iPhone,see http://www.apple.com/de/iphone/appstore/ 9 ForDBRailnavigator,see http://www.bahn.de/p/view/buchung/mobil/railnavigator.shtml

111 Although delays of long distance vehicles (e.g.trains) areconsidered, dynamictraffic data forshort distance (e.g.trams andunderground) is currently missing in the applications above. Whereas some services useGoogle Earthtodisplay thecurrent trip, neithertrue, dynamicdoor-to-door navigation, trip replanning,intermodalmobile ticketing,personalizationnor travel guidescan be foundinany of theservices so far. For thesystemtoplanthe trip in its entirety, includingtripreplanning,itisnecessary to have access to thetransportationcompanies’dynamic trafficdataonall themeans of transportationrequiredfor thetrip. In ordertorealizeconvenientand dynamic, real time door-to-door navigation in theabove-mentionedscenarios,the user’s positionhas to be continuously determined.Localisationisalsoimportant forstatusinformationregarding delays or time remainingtoreach thedesired meansoftransportation, forinstance, or for providingenvironment-specifictravelplans like historical buildings or cultural institutions.

3Location-BasedMobileTicketing

3.1Ring&Ride overview

An exampleofamobile ticketingproject whichuseslocationdatafor ticketing is the Ring&Ride10 project [Lu08].Incontrasttomostother mobile ticketing systems, Ring&Ride is basedonthe check-in/check-out concept, i.e. thecustomerhas to take an actionnot only at thebeginning, but also at theend of thetrip. When starting, he dialsa toll-free phonenumberand receivesaticket(SMS) that is valid forbothlongdistance (DeutscheBahn) as well as forshort distance travel.Atthe endofthe trip,the passenger dialsthe number againtosignalthatthe trip hasbeenfinished. Thecustomer’slocation is determined at thestartingpoint andatthe destination, butalsoduringthe trip at definedtimeintervals (cf. fig. 1).Therefore,the mobile phonehas to be switchedon during thewhole journeytoenablethe system to record thelocationinformation. After thecheck-out call, thelocationdataistransferredtoasubsystemcalled“routetracing”, whichhas access to infrastructure (public transportnetworks, i.e. busand trainstations, as ageo-codeddatabase) andtimetable data (schedule) andusesthe combinationofboth together with thelocationdatacollected to determinethe customer’s route.

Dependingonthe positioning accuracy,often notjustone,but severalpossibleroutesare found. Resultshaveshown that most of theroutes didnot differ much; forexample,they allmatched themeans of transportand lines availableand only differed in thestart or endstop, with acorresponding time shift[WS07].The last stepsinthe Ring&Ride processare to calculate thecorresponding fare andtosendthe customer an invoice.

10 Theproject wassupported by theGermanFederal Ministryfor Economicsand Technology andcarried out between2005 and2008byDeutscheTelekom AG together with theTechnical University Braunschweig,WVI GmbH,DeutscheBahnAGand otherpublictransportioncompanies basedinBerlin.

112 Figure 1: Locationdata(GSM) foratrip in Ring&Ride

TheRing&Ride idea hasthe following advantages:For customers, especially occasional andnew ones,itmeans fewerrestrictions andthe possibilitytotakeatripspontaneously withoutany knowledge of tariffsorticketingmachines. Ring&Ride also supports the idea of intermodalpublic transportacrossGermany.For thetransportcompanies,it wouldreduceticketdistributioncosts (atleast in thelongrun) by usingexisting infrastructure.

Twouserinterfacesweredeveloped: theinteractivevoice response system (IVR) interface can be used by everytypeofmobile phone. TheRing&Ride application(for: Java or WindowsMobile)additionallyprovidesroute andpricing information(afterthe trip).

In Ring&Ride,different locationtechnologies were used andcompared. It is examined howthe various locationmethods providedifferent qualitiesoflocationdataduringtrips with differentkinds of public transportvehiclesand at stations.Inparticular,the quality of thecalculatedroutesincreased significantly usingGPS or W-LANlocationmethods, even if they were only availablefor partsofthe route. Thesefindingscan be helpfulto optimizelocationmethods to supportall kindsoftravelsupportuse cases.

3.2Locationmethods fortravelscenarios

Thelocationsubsystem of theRing&Ride system,which is thecomponent responsible forcollectingand integratinglocationinformationfromdifferent sources,hides the varietyofpossiblelocationtechnologies from theother subsystems andusesthe most suitablemethodfor each situation. Alllocationinformationisstructuredinthe same way. A“position” consists of longitudeand latitude(thecentreofthe radiocell),the accuracy (the radius of thecell) andatimestampaswellasthe locationtechnology used. First,onlylocationdatafromGSM/UMTSmobile networks,providedbyDeutsche

113 Telekom’s PPGW11,was used.Inthe second step,GPS andW-LAN localisationwas added, butGSM wasstill used as afallback solution whenever theother location technologies failedorweretemporarilyunavailable. Thelocationsubsystem was designed to be flexible andextensiblefor newlocationtechnologies:itwould be possibletoadd GALILEO andNFC technologies without requiringmajor changes. Table1showshow thelocationtechnologies work in differentsituations on atrip. GSM/UMTS, GPSand W-LANweretestedinand around Berlin in Ring&Ride. The NFCtechnology is used in theTouch&Travelproject by DeutscheBahn[Ba09].

GSM/UMTS GPSW-LAN NFC Distributionof EverydeviceOnlymoderndevices Only modern Stillfew devices mobile devices devices Location cost Depends on Only fordata Only fordata Only fordata provider, contract transmission transmission transmission Quality of Cell radiuses vary High precision Exact position(for Exact position(for location from200 mto (distance<50m) fixedW-LAN fixedNFC terminal) severalkm router) Location at yesProblemsinbuildings Depends on NFCterminals have to stations andwithroofs existenceofW- be installed LANrouters Location on Mostly (some Big problems in trains No (evenifW- No (evenifNFC trains problems in becausemodernglass LANexistson terminal exists on tunnels) windows obstruct train, it moveswith train, it moveswith “view” of satellite train) train) Location on No12 No Like fortrainsLikefor trains underground Location on YesYes (mostly)Likefor trains Like fortrains buses,trams Table1:Comparisonofdifferent locationtechnologies

3.3Hybridpositioning

In theRing&Ride project,wedesignedand used hybrid positioning strategies,i.e.the locationsubsystem uses whatever method is applicable in thesituationtoget the customer’s location. We proposethe following ordertoproceed whenever the applicationneedslocationdata:

• Start with GPSdue to high precision.IfGPS is notavailable,try W-LAN. • If neitherGPS norW-LAN works(e.g.ontrain), useGSM as fallbacksolution. • NFCrequiresuserinteractionand theexistence of NFCterminals.Insituations whereacustomerusesNFC (likecheckinginfor atripwith Touch&Travel), this locationinformationshouldbeincludedaswelland override anyother informationbecauseofits high precision.

11 PPGWisDeutscheTelekom’s permission andprivacy gateway, atrusted thirdparty,neutral system platform whichprovidesinterfacesto4mobilenetwork operatorsinGermany forGSM location conforming to German permission andprivacy rules[EK06]. 12 Even if people can telephone,the locationmay be wrongdue to usageofrepeaters in underground stations.

114 4Infotainment/Community

Thesituationofpublic transporttravellers offers auniqueopportunity to supplythem with information, entertainmentoradvertisements.Duringthe trip,manytravellers are idle forsometimeand responsiveto”kill-time“ offers.Withtravelinformationathand, theirsituationcan be characterisedveryprecisely with respect to time andlocation. Furthermore, personal preferences couldbeobtainedthrough theintegrationof community systemssuchasFacebook.Thisenables thetransportationcompanies to providetheir customerswithspecially-tailoredinfotainmentoffers.

Both data storagesystems andcontent arealreadyavailableinestablished andwidely- used systems. DB AG,for example, alreadyoffers theircustomers apersonalised system for travel information.Hence,detailedinformationabout thetravelscheduleofa particular customer can be made available.

Many different content storagesystems areathandand areusedbymobile stand-alone applications to retrieve location-specificinformation.Examples areAroundMe, WikiMe etc. forthe AppleiPhone. Personal preferences that go beyond travel-specificdataare maintained by an increasingnumberofpeoplethrough diversecommunity systemssuch as Facebook,Xing, LinkedIn etc. We proposetoprovide themissing link betweenthese systemsbyaRESTfularchitecture(cf.section5)thatbringsthe currently available informationtogetherand easetoprovide user interfacesfor thediverse mobile platforms.

Community supportisnot limitedtouse of existingdata. On thecontrary, we seeatleast twointerestingforms of meaningful data generationcausedbythe travel:usergenerated travel informationand automatedpresenceinformation. User generatedcontent anduser generatedlinks to existing contentcan be exploitedtoprovide an increasinglyrichand valuable setoflocation-specificoffers.Geoinformationsystems such as Google maps arealreadyconnected to locationspecificcontent such as user generatedphotosor mixed-mediacontent in wikipedia. Thereisnobig technical barrier to deliver multi- mediacontent aboutacustomer‘s routeordestinationwhile sheistravelling.

Foragrowingnumberofpeoplepresenceindiverse communitysystemisadaily necessity.For microbloggers (suchasTwitterusers)the situationoftravelling is certainly worthtoinformfriends aboutit. Butalsomoreserious usersofcommunity systemssuchasXingare probablyinterestedinupdatingtheir status online.An integrated travel system as we proposeitinthispaper couldsupportpersonalised presence informationbyautomated status updates.

5Towards an Architecturefor Integrated Travel Services

Fig.2showshow trip planning,mobile ticketingand thirdparty services as infotainment andcommunity supportcan be integrated to yieldaconvenientand coherent servicefor travellers.Weillustrate aproposed system architecturebyfocusingonfour usecases: plan trip, replan trip, select infotainment,and participateincommunity.

115 We emphasisetwo requirements here:(1) boththe trafficdataservice andall involved thirdparty services must be easily accessible andintegrableand (2)diverse mobile client platformsmustbeuseable forthe integrated serviceset.

State is crucialfor trip planning in termsoftimelinessoftransportand resulting dynamic expected arrivaltime.13 Thedatabehindatrip plan changesdynamically during travel, e.g.atraingetsdelayed so that acertain connectingtrain will be missed.With the commonsession-basedimplementations (e.g.ofthe DB), trip plansare notpersisted server-side. Newtravelsituations can notbedetected automatically,but mustbe perceivedbythe customer whomanuallyhas to createanew trip plan.Bettersupporton theclient side is difficulttoimplement (poll) andveryspecial foreach client platform.

Hence, we suggesttoreify thecustomer‘sindividualtripplanasaserver-sideresource as listing 1exemplifies.Wepreferastatelessclient server communicationfor alltrip planning functions to thecommonsession-basedconnections.The stateofthe trip plan is thereforemoved from theapplicationtothe resources.

1(2+0'"(3-(04 5680+3-! !"# %&'(("() *+,"&- "./-0"() 5+336("07

97('3". 0!'22".:'0' 5!-'0- 0!"# %&'( 0!"# #&'(

5!-'0- ('=")'0"+( ;+.'0"+( #&'(

<'=")'0-0+ 80'0"+( 56!!-(0 0!"# #&'(

5!-'0- *+,"&- .>-./?1( 5>-./?1( :'0' 1(2+0'"(3-(0 :'0'

@-#&'( !-.'&.6&'0- 0!"# 0!"##&'( 5!-'0- .>-./?B60 :'0' %!+=":- A-&-.0 .680+3-! 1(2+0'"(3-(0 8#-."2". "(2+0'"(3-(0

E..-88 0+ %'!0"."#'0- 8+."'& "( .+336("07 .+336("0"-8 !'.-!+60- C 5'&.6&'0- 0'!"22 *+,"&- A+."'& 5>-./B60 (-0D+!/8

Figure 2: Illustration of trip planning,mobile ticketingand thirdparty services

13 E.g. userscan keep each otherinformedabout thestatusoftheir trip.Aswellasproviding navigation informationincars, advanced TomTom navigationsystems have aconnectiontocellularphonesand cansend informationtoTomTomservers when acar hasstopped on thehighway.Livetrafficinformation (when multiple messagesfromthe same road have to come within acertain timeframe)will suggest rerouting.

116 Oursuggested architecturefollows theREST14 style[RiR07],i.e.each resource is addressablebyanURI andaccessedvia thestandardhttp methods GET, POST,PUT andDELETE. Resources arelinkedwitheach other. E.g.the XML sample foratrip plan resource links to otherresources likestations,trainsand theuser‘sfinal destination.

In fig. 3wesketchthe suggestedresource-oriented architecture. Theresources are presentedanaloguetoUML classes, e.g. /tripplans represents theset of allcurrently stored trip plansonthe travel management server while /tripplans/{id} represents an individual trip plan as exemplifiedinlisting 1(with id =0815).Each resource can be accessedbythe standard http methods.E.g.calling POST with an appropriate setof parameters on /tripplans creates anew individual trip plan.Later on during the travel,thisresourcemight be updatedwithdynamictraffic data or user triggered replanning by usingthe PUTmethodwithappropriate parameters.

Figure3:Sketchofresourceorientedarchitecture

14 REST meansrepresentationalstate transfer andhas evolvedtoafirm setofcharacteristics easing web- serviceinteraction.

117 Hence, thetripplanispersisted on the server andaccessiblefor readingand 0815 updatingbysimplyusing theresource bahn.de/tripplans/0815/destination URIand http methods.All client-server communicationisatomicand state-less, bahn.de/tripplans/0815/idletime so no session time outs will occur. Furthermorethe direct accessto Köln Hbf resources provides aseamless bahn.de/stations/koeln integrationwithfurther informationof 2009.02.21_05:49 thetransportcompany.The trip plan is linkedwith trainand station Berlin Hbf bahn.de/stations/berlin informationand thelikeasshown in listing 1. To increaseinteroperability, 2009.02.21_10:11 special resourcescan extend thecore ICE853 set(cf.fig.3). Forexample,the time bahn.de/trains/ICE853 until thetravellerneedstotakeaction again, e.g.todisembark,ismodelledas aresourceonits own. Calling GETon theresource /tripplans/0815/ idletime retrievesthe current idle time of theuser. This allows athird Listing1:AbridgedTripPlanasXML partyservice to search automatically formovieswithamaximumplaytimebysimplygivingthe URItothe idletime resource as search parameter. That couldeasilybeintegratedintoahtml page of thepublic transportcompany.

Resourcescan be representedindifferent formats, dependingonthe clients’ demands. Hence, thesameresources canbeusedbyawebbrowser (receiving html)and e.g. an iPhone application(receiving xml).The selectionisdoneusing thestandardacceptmeta data of thehttpcall.

6Outlook

In this paper, we show that integrated solutionsinpublic transportconsistingoftrip planning,location-basedmobile ticketing andinfotainment/community services can enhancethe customer´s satisfaction. We sketched an architecturehow thirdparty services (e.g.community andinfotainment) can be integrated.Inorder to continue with theseideas,the proposed architectureneedstobeadjustedtoexistingsolutions (e.g. mobile ticketing).Inthe future,aprototypethatcoverstripplanning,mobile ticketing andinfotainmentaswellascommunity topics will be setupand implementedat different public transportationcompanies.

118 Theintegrationofreal-timeinformation, infrastructure data andactualtimetable data will also be anecessary task,considering that “single-sign-on” functionalities(e.g. severalpublic transportationcompanies)willenhance theusabilityand thecomfort of thesolutionproposed above. Additionally,primary research forcustomerevaluationand acceptance testsshouldbeconsideredinfurther work.Furthermore,businessmodels have to be analysed anddifferent end-user devices have to be considered.

References

[Ba09] Deutsche Bahn: Websideofhttp://www.touchandtravel.de/,2009

[Bi08] Berg Insight: Mobile LocationBased Services,2008

[Ca09] Canalys: Website http://www.canalys.com/pr/2008/r2008112.htm,2009

[Ca07] Canalys: SmartMobile Device andNavigationTrends 2007/2008, 2007

[EK06] Eichler,G.; Karge, R.:ALocationand PrivacyMiddlewarefor Context-awareMobile Applications,ICIN,Convergence in Services,Media andNetworks, Bordeaux, 2006

[Fr06] Forrester Research:GettingConsumerstouse Mobile Services,2006

[Lu08] Lüke,K.-H.;Schlüter, W.;Telschow,A.; Sommer,C.; Bley,O.; Wermuth, M.: Location-BasedMobile Ticketingwith ElectronicFareManagement, IPTS conference proceedings,Amsterdam,2008

[MLE07] Mügge,H.; Lüke,K.-H.;Eisemann, M.:Potentials andRequirements of Mobile Ubiquitous Computingfor Public Transport, in:Gesellschaftfür Informatik (GI) Proceedings 110, Band 2(ISSN 1617-5468),Bremen, 2007

[RiR07]Richardson, L.;Ruby, S.: RESTfulWeb Services,O‘Reilly2007

[Sa07] Strategy Analytics: Consumer LocationBased Service, 2007

[Vi08] Visiongain:Mobile Social Networking &Usergenerated Content, 2008

[WS07] Wermuth, M.;Sommer,C.: Neuere Entwicklungen beim Handy-Ticketing. In:Der Nahverkehr 7-8/2007.P.51-55.Düsseldorf: Alba Fachverlag GmbH Co.KG, 2007

[Zu05] Zumkeller, D.;Manz, W.;Last, J.;Chlond, B.:Die intermodale Vernetzung von Personenverkehrsmitteln unterBerücksichtigung derNutzerbedürfnisse (INVERMO) Schlussberícht zumBMBFProjekt 19 M9832A0, Karlsruhe,Germany,2005

119

Session4

Web Portalsand Usability

AHybrid Approach to Identifying User Interests in Web Portals

Fedor Bakalov1,Birgitta Konig-Ries¨ 1,Andreas Nauerz2,and Martin Welsch2

1Friedrich Schiller University of Jena, Institute of Computer Science Ernst-Abbe-Platz 1-4, 07743 Jena, Germany {fedor.bakalov|birgitta.koenig-ries}@uni-jena.de

2IBM Research and Development Schonaicher¨ Str.220, 71032 Boblingen,¨ Germany {andreas.nauerz|martin.welsch}@de.ibm.com

Abstract: Webportals pioneered as one of the earliest adopters of personalization techniques to help users dealing with the problem of information overload. Nowa- days theyare extensively used as asingle-point of access to the vast amount of resources available on the Weband in enterprise intranets. Anumber of researchers have been investigating the possibilities to enable portals to deliverthe content in a highly-personalized manner in order to provide users with aquick and efﬁcient access to the subset of resources relevant to their information needs. However, in order to achieve such apersonalization effect, the portal needs accurate and up-to-date information about users, especially the information about their interests. In this paper,we describe ahybrid approach to identifying user interests in Webportals. In our approach, the portal is enabled to “learn” the user interests from the content of visited pages. In addition, it is empowered to provide users with an open access interface to their user models to let them explicitly specify their interests and, in case of incorrectly identiﬁed interests, outvote the portal.

1Introduction

Webportals emerged in the late 1990s primarily as the gateways to different information resources available on the Internet. Companies likeAOL, Altavista, and Yahoo used them to guide their user communities through the network. The first portals provided users with Webdirectories, fulltext search capabilities, and certain communication services like email and chat. Later,portals gained special attention among enterprises as aplatform to integrate not only the corporate information resources, butalso the company’slegacy systems. Nowadays, alarge number of organizations use portals extensively as acomplete e-business solution providing users asingle-point of access to the vast amount of company resources and applications. Furthermore, with the advent of Web2.0, portals have gained popularity as the gateways to community-drivenresources likewikies, blogs, mashups, and manyothers. However, with the constantly expanding growth of information available through Web portals, it has become more difficult and time consuming to find relevant resources. In

123 today’sWeb 2.0 this problem has become even more prominent due to the large number of different users contributing. In case of aportal consisting of several hundred pages with contributions from different members of acommunity,traditional hierarchical navigation is no longer efficient as it is not feasible for the administrator to come up with one optimal topology that would fit to the navigation patterns of every individual user.Furthermore, it is unlikely that users will place the content that theycontribute according to some plan an administrator drewup. Thus, the portal industry is facing achallenging research question -how to makeportals adaptive to the information needs of individual users. Anumber of researchers have been investigating the possibilities to createindividual personalized information spaces, where the user could easily access the subset of resources that are relevant to his/her information needs [GAP07, CGP07, BKA98]. However, in order to create such apersonalized information space, the portal needs to “know” certain information about the user.Such information is usually stored in asocalled user model. Brusilovskyand Millan´ defined user model as “a representation of information about an individual user that is essential for an adaptive system to provide the adaptation effect, i.e., to behave differently for different users” [BM07]. Depending on the desired adaptation effect, the user model may contain such information as user interests, background, goals, traits, etc. In this paper,wedescribe ahybrid approach to identifying user interests in Webportals. The goal of our research is to enable the portal to “learn” the user interests from the content of visited pages. In addition, we aimed to provide users with an open access interface to their user models to let them explicitly specify their interests and, in case of incorrectly identified interests, outvote the portal. The remainder of this paper is organized as follows. Section 2and Section 3provide details on the user model and domain model respectively. In Section 4, we elaborate on the approach itself and describe the steps involved in the user modeling process. In Section 5, we describe the prototypical implementation of our approach and provide results of apreliminary evaluation. Section 6provides ashort overview of the previous research that has been done in the area of user modeling. Finally,Section 7concludes the paper and outlines the directions for our future work.

2User Model

In our approach, the user model logically consists of twoparts: astatic part andadynamic one. The static part contains the time invariant information about users: date of birth, gender,ﬁrst language, etc. This is the information that users provide explicitly when they register in the portal. In the dynamic part, we represent the constantly changing user features, namely user interests. Basically,wedeﬁne user interest as afact indicating that agiven user is interested in a certain term with acertain degree of interest. Here, the term is areference to aconcept denoting either areal world object, likecompany, geographic location, person, etc., or an abstract notion, likearea of science, discipline, technology,etc. The concepts themselves are stored in the underlying domain model,which is represented as an ontology providing

124 machine-processable semantics of the contained entities (see Section 3). The degree of interest denotes the extent to which the user is interested in agiven term. We distinguish three levels of interest and identify them by the following linguistic variables: interested, partially interested,and not interested.Also we introduce an auxiliary linguistic variable blocked to mark the terms that were explicitly blocked by the user.This information could be used by the portal to stop tracking user interest in the giventerm as well as by the personalization engine to stop recommending resources about that term. Section 4.3 explains in detail the methods for determining degree of interest. Following the research described in [Sch06], we model user interests as time dependent features. We assume that auser might be interested in acertain term only for acertain period of time. Forinstance, afootball fanisprobably interested in the World Cup mostly during the time period when the event takes place. Thus, the interest user model is represented as acollection of tuples (U,T,I,V),where:

• U is the user portal ID • T is the URI of an instance from the domain model • I is the linguistic variable indicating the degree of interest • V is the time period of the interest validity

Forinstance, the fact denoting an interest of aﬁctional user Klaus in the World Cup 2010 can be represented as: (klaus, http://www.minerva-portals.de/domain-model.owl#World Cup2010, interested, (2010-06-01, 2010-07-31)).

3Domain Model

Adomain model is adata model that defines concepts in agiven domain, e.g., chemistry, medicine, biology,etc. [GPFLC03]. We have chosen the finance domain for our proof- of-concept implementation. Therefore, in our domain model we define the concepts that users from the financial realm may work with, such concepts as stock, bank, account, etc. The domain model is represented as an OWLontology1 (Figure 1), which defines the domain concepts as ontological classes by specifying their properties and relations to other classes. E.g., an event Acquisition,denoting the fact of acquiring one companybyanother one, is defined as asubconcept of FinancialTransaction and is described by such attributes as date, acquiree, acquirer,and transationAmount.The ontology is grounded on the Pro- ton Upper Module2,which defines upper-levelconcepts, such as organization, location, person, etc. Under these concepts, we define finance-specific terms partially reused from twoexisting finance ontologies, namely LSDIS Finance Ontology3 and XBRL Ontology4.

1http://www.minerva-portals.de/ﬁnance-domain.owl 2http://proton.semanticweb.org/ 3http://lsdis.cs.uga.edu/library/resources/ 4http://xbrlontology.com/

125 Figure 1: Domain Model

The domain model also contains instances of concepts. E.g., for the concept Company,we specify such instances as “Microsoft”, “IBM”, “Google”, etc. Class instances are required to represent speciﬁc user interests in the user model. Inclusion of newinstances into the ontology is performed automatically using the Calais service (see Section 4.1). We harness the service in order to extract named entities (such as company, industry term, technology, etc.) from the text of documents accessed by users. The extracted entities that are not yet present in the domain model are then inserted into the ontology as instances of the corresponding concepts.

4Approach

Our approach to identifying user interests involves the following activities. First, the portal collects the terms indicating user interests into the user model by analyzing the content

126 of visited pages as well as by allowing users to explicitly enter them through aspecial interface. Second, the collected terms are semantically enriched by refering to the corresponding instances in the underlying domain model. Finally,for every collected term, the portal determines degree of interest either by leveraging the term frequency, or semantic relation among the terms, or by letting the user specify it explicitly.

4.1 Collecting Terms

In our approach, we distinguish twosources for collecting newterms into the user model. First, we allowfor automatic extraction of terms from the pages visited by the user.Sec- ond, we provide the user apossibility to manually enter newterms into his/her user model. Forthe automatic extraction of terms, we leverage the Calais Webservice5.Calais is an unstructured text analysis service, which can receive an HTML or plain text document as an input and return an annotated document in RDF6 format. More specifically,the service performs named entity recognition: it extracts certain general and business-related entities such as company, location, person, etc. It also supports extractionofcertain events and facts, such as acquisition, bankruptcy, family relation, etc. We developed aspecial portlet that analyzes the content of pages visited by the user with the Calais service and uses the extracted entities to update the domain and user models. Forevery extracted entity,the portlet checks if the domain model contains amatching instance. If it does not find one, it inserts anew domain instance using the information about the extracted entity provided by the service, which includes the entity label, full name, and type. Afterwards, the portlet makes an entry in the user log where it specifies the URI of the domain instance and the number of occurrences in the document. Finally,the portlet determines the degree of interest in the term (see Section 4.3) and inserts anew user interest into the user model. In addition to the automatic extraction of terms, we empower users to manually enter new terms. We developed auser modeling portlet (see Section 5) where the user can access the terms stored in the underlying domain model and insert the terms of interest into his/her user model.

4.2 Determining Semantics of Terms

As mentioned above,the semantics of the existing user interests is represented in the underlying domain model. In case when the user manually inserts anew interest, the portal already “knows” its semantics because it is an existing domain instance. In the current implementation of the approach, we allowusers to select newinterests only out of the existing domain instances. However, we are currently investigating possibilities to enable user communities to collaboratively edit the portal domain model by adding newconcepts and instances as well as connecting them through user-deﬁned relations. Thus in the future, we plan to enable users to specify newinterests by adding anew ontological instances

5http://www.opencalais.com/ 6http://www.w3.org/RDF/

127 Figure 2: Fuzzy sets representing degree of interest directly into the domain model. In case of newterms extracted by the Calais service, the portal needs additional information to determine the semantics. Forthis purpose, we developed amapping between the Calais types and the concepts stored in the domain model. Forevery Calais type, we identiﬁed acorresponding domain concept and mapped it to the Calais’ one through the OWL sameAs property.For instance, domain concept StockIndex is mapped to the Calais type MarketIndex through the following assertion: http://www.minerva-portals.de/ﬁnance- domain.owl#StockIndexowl:sameAs http://s.opencalais.com/1/type/em/e/MarketIndex.All newterms coming from the Calais service are inserted into the domain model as instances, which classes are determined using the mapping. E.g., using the above mentioned mapping, term FTSE 1007 will be inserted as an instance for the domain concept StockIndex.

4.3 Determining Degree of Interest

As described in Section 2, we distinguish three levels of user interest and identify them by the following linguistic variables: interested, partially interested,and not interested.Every variable is associated with afuzzy set, which is deﬁned by the corresponding membership function [Zad65, Kav04]. The membership functions are based on cumulative weight,a real number that can takevalues from 0to1,that denotes importance of acertain term for the user with respect to the other terms. We use this value to deﬁne the membership functions that represent the degree of user interest; µni, µpi, µi showthe degree to which the user is not interested, partially interested,and interested in the giventerm (Figure 2). The cumulative weight is aconstantly changing value and can be determined with one of the following three methods: log-based updates, inference-based updates, and user manual updates.

7Financial Times Stock Exchange Index

128 Figure 3: User modeling portlet

Log-based updates are performed automatically by the portal using the user log that stores occurrences of the terms extracted from visited pages. Forevery term in the user model, the portal calculates the term frequencyvalue:

ti,j TFi,j =! tk,j k where t is the number of occurrences of termi for userj,and the denominator is the total number of occurrences of all terms registered for userj.The computed term frequency value of termi is specified as the term’scumulative weight, unless the user has already manually specified his/her interest in the term. Inference-based updates leverage the semantic relations among the instances in the domain model. This is when the portal identifies newinterests by propagating interest from the terms for which the user model already contains information about (e.g. the user has explicitly specified interest degree or it has been determined by the portal based on the term frequency). Forinstance, if the user model contains auser interest in Berlin and in the domain model Germanyisconnected to Berlin through the property hasCapital, then the inference engine can propagate user interest from the former to the latter.The taxonomical relations can be leveraged for interest propagation as well: interest can be propagated from achild concept to its parent and vice versa. User manual updates are performed by the user explicitly.Wedeveloped auser mod-

129 eling portlet (displayed in Figure 3) through which the user can access and edit his/her user model. The user can check the interest status generated by the portal and in case of incorrectly identiﬁed interest, the user can outvote the portal by manually changing the status, which will also affect the cumulative weight value. If the user promotes aterm (e.g. from not interested to interested), the cumulative weight of that term will be increased up to the lowest cumulative weight value in the fuzzy set of “interesting” terms. Whereas, if the user demotes aterms (e.g. from interested to not interested), the cumulative weight will be set as the highest cumulative weight in the fuzzy set of “not interesting” terms. In case the user blocks aterm, the cumulative weight of that term will be set to 0.

5Implementation and Preliminary Evaluation

The approach described above has been prototypically implemented in IBM’sWebSphere Portal. Figure 4illustrates the system architecture of the prototype. The portlet application consists of three portlets. Content portlet provides users with news harvested via RSS feeds from news websites likeBBC and CNN. The pages displayed in the content portlet are processed with the models update portlet,which sends the content to the Calais service and based on the extracted entities updates the user and domain models. Finally,the user model portlet,displayed in Figure 3, allows users to viewcontent of their user models, add newinterests, and change interest status of existing terms. Additionally,the portlet allows users to reorder the terms within the same interest group using the drag-and-drop technique.

Figure 4: System architecture

The user model is implemented as arelational database. It stores information about user interests and logs containing the user browsing history and user model updates. The domain model is implemented as an RDF triple store deployed in the Sesame Framework8. Access to the user model and domain model is provided through user modeling service and domain modeling service respectively.The user modeling service provides such operations as, adding newlog entry,performing log-based updates and user manual updates, and providing user interest list. Whereas, the domain modeling service is used for adding

8http://www.openrdf.org/

130 newdomain instances and getting information about the existing instances, such as main label, aliases, and class. The prototype has been evaluated with three ﬁctional users representing three countries, namely,Germany, Russia, and France. Every user used the content portlet to read news related to the country he or she is representing. The content of visited pages wasprocessed with the models update portlet,which created three corresponding user models. Table 1 shows the results of experiment.

User News Visited Total Interests Topic Pages Terms Interested Part. Interested Not Interested Klaus Germany381 638 530603 Dmitry Russia 72 226 859159 Isabelle France 14 45 28 17 0

Table 1: Results of experiment

We compared the three generated user models and identified the following issues. As the reader can see from Figure 5, the proportion of interests depends on the size of the model. Here the user model of Klaus,the largest model, contains less than 1% terms that the user is interested in, whereas in the user model of Isabelle,the smallest model, the number of terms that the user is interested in exceeds 62%. This happened because the functions determining the membership of aterm in the interest fuzzy sets are based on the cumulative weight value, which is by default set to the term frequency, the value that depends on the size of the term set. Therefore, the membership functions must be defined for every user individually based on the size of the corresponding user model. We are currently investigating possibilities to makethe fuzzy sets floating and adaptive to the size and content of the individual user models. We also aim to enable users to manually adjust the fuzzy sets, by which theycan increase or decrease the number of terms in acertain interest group.

6Related Work

The user model is an essential component for anysystem that aims at adapting content to the users’ speciﬁc needs. Anumber of proposals have been made to identify and represent knowledge about users [BGT87, Kob01, Fis01]. Crabtree and Soltysiak [CS98] describe an approach to deriving user interests automatically by monitoring various user activities, such as reading documents, writing emails, and browsing websites. The resulting user model is represented as avector of weighted keywords denoting user interests. In [AHNJ07], Achananuparp et al. describe howthe vector user models can be semantically enhanced. Theypropose using WordNet9 lexical database to establish the semantic relations between the keywords in the user model. Ontologies have gained alarge interest as ameans to semantically represent knowledge about users. Semantic relations among concepts in the user model allowderiving cer-

9http://wordnet.princeton.edu/

131 Figure 5: User interest statistics tain information about users that wasnot explicitly deﬁned in the model. Forinstance, ontological representation of user interests enables propagation of the interest from child concepts to their parents as well as among similar concepts. In [Sch08], Schmidt describes ontology-based conceptual models that can be harnessed by learning management systems in order to provide users with the learning content tailored to the leveloftheir knowledge as well as their situational needs. In [ZSS07], the authors elaborate an architectural solution that can automatically derive ontological user models based on Webcontent and user logs. Asimilar approach is described by Costa Pereira and Tettamanzi in [PT03]. An important aspect of user modeling is the ability to identify and represent the degree that the user is interested in acertain concept or possess knowledge on it, which in its turn can affect the quality of adaptation. In [Kav04], Kavcic describes anovelapproach to representing the degree of user knowledge using fuzzy set theory.John and Mooney describe asimilar approach to represent user interests in [JM01].

7Conclusions and FutureWork

In this paper we have presented ahybrid approach to identifying user interests in Web portals. We empowered the portal to harvest user interests in anon-intrusive manner by analyzing the content of visited pages and extracting semantic entities from them, which are then used to update the user interest models. Additionally,wehaveproposed an open interface to let the users explicitly specify their interests and, in case of incorrectly iden-

132 tiﬁed interests, outvote the portal. Our approach has been prototypically implemented in IBM’sWebSphere Portal. The information contained in the user model can be used to adapt the portal to user needs. We have showcased this with aless sophisticated user model in our previous work [NBKRW08]. In the future, we intend to build on that work to provide truly personalized portals. We will also extend the user model proposed here by leveraging community interests as well as individual user’sinterest. Beyond that, we will focus on evaluation of our ideas in areal-world setting.

Acknowledgements and Trademarks

This work has been done in the framework of the MinervaPortals Project funded by IBM Deutschland Research &Development GmbH. IBM and WebSphere are trademarks of International Business Machines Corporation in the United States, other countries or both. Other company, product and service names may be trademarks or service marks of others.

References

[AHNJ07] Palakorn Achananuparp, Hyoil Han, OlfaNasraoui, and Roberta Johnson. Semantically enhanced user modeling. In SAC’07: Proceedings of the 2007 ACMsymposium on Applied computing,pages 1335–1339, NewYork, NY,USA, 2007. ACM. [BGT87] Giorgio Brajnik, Giovanni Guida, and Carlo Tasso. User modeling in intelligent information retrieval. Information Processing and Management,23(4):305–320, 1987. [BKA98] Krishna Bharat, Tomonari Kamba, and Michael Albers. Personalized, interactive news on the Web. Multimedia Systems,6(5):349–358, 1998. [BM07] Peter Brusilovskyand EvaMillan.` User Models for Adaptive Hypermedia and Adaptive Educational Systems. In Peter Brusilovsky, Alfred Kobsa, and Wolfgang Nejdl, editors, The Adaptive Web: Methods and Strategies of WebPersonalization,volume 4321 of Lecture Notes in Computer Science,chapter 1, pages 3–53. Springer,Berlin, Heidelberg, 2007. [CGP07] Alison Cawsey, Floriana Grasso, and Cecile´ Paris. Adaptive Information for Consumers of Healthcare. In The Adaptive Web, Methods and Strategies of WebPersonalization,pages 465–484, 2007. [CS98] Barry Crabtree and Stuart Soltysiak. Identifying and tracking changing interests. Interna- tional Journal on Digital Libraries,2(1):38–53, October 1998. [Fis01] Gerhard Fischer.User Modeling in Human-Computer Interaction. User Modeling and User- Adapted Interaction,11:1:65–86, 2001. [GAP07] Anna Goy, Liliana Ardissono, and Giovanna Petrone. Personalization in E-Commerce Ap- plications. In The Adaptive Web, Methods and Strategies of WebPersonalization,pages 485–520, 2007. [GPFLC03] Asuncion´ Gomez-P´ erez,´ Marianno Fernandez-L´ opez,´ and Oscar Corcho. Ontological Engi- neering.Advanced Information and Knowlege Processing. Springer,2003. [JM01] R. I. John and G. J. Mooney. Fuzzy user modeling for information retrievalonthe World Wide Web. Knowledgeand Information Systems,3(1):81–95, 2001. [Kav04] Alenka Kavcic. Fuzzy user modeling for adaptation in educational hypermedia. Systems, Man, and Cybernetics, Part C: Applications and Reviews, IEEE Transactions on,34(4):439– 449, Nov. 2004.

133 [Kob01] Alfred Kobsa. Generic User Modeling Systems. User Modeling and User-Adapted Interac- tion,11(1-2):49–63, 2001. [NBKRW08] Andreas Nauerz, Fedor Bakalov, Birgitta Konig-Ries,¨ and Martin Welsch. Personalized recommendation of related content based on automatic metadata extraction. In Marsha Chechik, Mark Vigder,and Darlene Stewart, editors, CASCON,page 5. IBM, 2008. [PT03] Ce’lia Da Costa Pereira and Andrea Tettamanzi. An Evolutionary Approach to Ontology- Based User Model Acquisition. In Vito Di Gesu‘, Francesco Masulli, and Alfredo Petrosino, editors, WILF,volume 2955 of LectureNotes in Computer Science,pages 25–32. Springer, 2003. [Sch06] Andreas Schmidt. Ontology-based user context management: The challenges of dynamics and imperfection. In On the Move to Meaningful Internet Systems 2006: CoopIS, DOA, GADA, and ODBASE. Part I., ser.Lecture,pages 995–1011. Springer,2006. [Sch08] Andreas Schmidt. Enabling Learning on Demand in Semantic Work Environments: The Learning in Process Approach. In Jor¨ gRech, Bjorn¨ Decker,and Eric Ras, editors, Emerging Technologies for Semantic Work Environments: Techniques, Methods, and Applications.IGI Publishing, 2008. [Zad65] L. A. Zadeh. Fuzzy sets. Information and Control,8(3):338–353, June 1965. [ZSS07] Hui Zhang, Yu Song, and Han-Tao Song. Construction of Ontology-Based User Model for WebPersonalization. In User Modeling,pages 67–76, 2007.

134 Web2.0 as an autopoieticsystem -implications for innovative web-interfaces -

Kathrin Vent

Deutsche Telekom Laboratories Ernst-Reuter-Platz 7 10587Berlin [email protected]

Abstract: The Web 2.0 can be regarded as an evolutionaryprocess of medial differentiation.Infiniteonline communities areemerging and disappearing.It seems that arace has started in searching for innovativeweb-interfaces. Already manyhelpfultechnologicalprinciples and features are executed by web experts. Although they give helpfulorientation for designinginnovativeand interactive web-interfaces, theylackdealingwith self-organising systems suchasonline communities. Therefore,Ipromote an extended notion of the web-interface. Such perspective includes reconsideringthe constitutive characteristic of online communities: theautopoiesis. This is an essentialconceptofsocialsystemtheory, which is new in contextofdesigning innovative web-interfaces. Consequently,the goal of this paper is not atechnologicalsolution, but which will give anew perspectivewithimplications for technology. Oneessential consequence is that a designer or programmer cannotbemore than aperturbator and experimentator – intendingtolearnmore about thesocial system. Therefore, Ipropose to generate knowledge for an innovativeand human centred Web2.0 interface with focus on qualitativeand participatoryresearch.Thisapproach allows generating most appropriatedesign interventions with the users of an online community themselves.

Keyword: communication, media, web 2.0, social systems, autopoiesis, emergence, extended interface notion, qualitativeresearch

1Background

1.1The evolutionofCommunication Media

Internet in general,and Web2.0 in particular,are often consideredtoberevolutionary in termsoftechnology-assisted human communication.Althoughnoone will arguethe unprecedented opportunities these technologies provide, ashiftofperspective/language fromrevolution to evolution,inour opinions,will bring valuable implications for the web interface design andbeyond.

135 First of all, since interactiveweb-services have already existedbefore, likewikis,blogs or even emails and chat,itmight be more appropriate to describe theinnovative development of the Webasanevolutionaryprocess.Krippendorff highlightsthatthe evolutionofmedia does not primarily base on technology (Krippendorff, 94)Rather, it startswith thebigger,perhaps thegreatest human“invention” of communicationmedia, namely, language, scriptsand printing. These should also be considered “invented” mediaand theevolutionary driver of thedifferentiationofcommunicationsystems.

Onemight still ask if theWeb 2.0 is notarevolutionbut merely asmall step in aslow evolutionary process, what makes it so popular? Themain reason,aswell known, is that it allowsmultiplewaysofcommunicationacross physical or cultural boundaries. This enables already existing communication patterns to appear in anew form–more distinctiveand differentiating. Thecase of collectiveintelligence, also known as emergence canserve as an example. James Surowiecki (2005)pointed outthatthe cognitive efforts fromasizable butordinarygroup of people, set up arbitrarily, can reach ahigher intelligence level than asmall groupofexperts. Themain determiner is the social emergence, whichdevelops itselfthrough interactiveand individual groupwork. Asuitableanalogy of such functioning of social systemsisant colonies.Although a single ant does not get hierarchic instructions,theyare well organized. Decisive foran antcolonyand aweb serviceisthatthe social system can auto-organise through participation andcollaborationamong individuals. However, agroupalone is not sufficient to create collective intelligence; preconditions are technologicalsystems, like thewell-citedWikis. These technical systems enable acommunicativepraxisofsocial systems,constituted and confirmed by different perspectives and experiences (Willke, 2002).

Therefore, theevolutionofthe webisbetterregardedasaself-energising process of mediadifferentiation, ratherthananabruptrevolution.More communicationneeds provoke- more mediadevelopment and more mediaprovokes exponentially more communication. However, social systemslikeanonlinecommunity not onlyemerge through communication –but also ceasetoexistwithout it. If anew onlinephoto sharing community having thesamedesign andtechnologyasFlickr couldnot attract users who publishphotos, thecommunicationonthe platform woulddie.The online community wouldvanishlikesomanyother internet start-ups.Evenwithsophisticated technology, it is useful to remind that onlinecommunities are disappearing as fast and as often as emerging.

1.2Problem statementand value

After thefirst dotcomcrisis, it hasbecome clear that technical advancement on Web2.0- application alonewas notsufficienttosustainon-linecommunities or thegrowth of Web 2.0. (O`Reilly 05). Since then,many have turnedattention to innovative web interfaces. Interfaces of the most successful online communities like eBay, Xing andYouTube are analysed, discussed andrecommended.

136 Arace seemstohave started to develop principles to helpand to instructhow to deal with web technology. For instance, theprinciple: “End of asoftware release cycle” (O`Reilly 05) which suggests that software is not anymoreoffered as afinished product, but it getsrenewed andactualiseddaily, depending on usage frequency.Thisprinciple might help,but whenwelook more closely, we realise that thefrequency of usages does not tell us anythingaboutthe motivationofthe user.Maybe thegrantedfeatureswere more comprehensiblethanothers were; maybethere were notother feature what could express better theindividualrequest until now.

Socialsystems do not emerge primary from technical framework but are the presuppositionfor alreadyexistingneeds in society. Pets.com for example invested millions euro in an onlineshopsystem to sell cat litter; as theproductdid not stimulate demandonthe market,the company went bankrupt.

It appears,oncereachedall this technological expertise, to be able to rule over aself- created empire (web-platform)with amodern infrastructure (principles). Butthe technical basis to build up aplatform that is notsufficient. That mainly lies in thefact that the notion oft theinterfaces leaves out asocial and human centred designing components.

2Social system theory: Anew approach for intervention

2.1Anew perspective

Technological features andprinciples areessentialfor designing innovativeweb- applications, butnot acondition for an ongoing communication and therefore successes of aplatform.Theylack dealingwith theself-organising systemofonline-communities. Therefore, Ipropose an extendednotionofweb-interfacesthatincludes reconsidering theconstitutivecharacteristics of an online-community: theautopoiesis.

Thenotion of theautopoiesis (fromGreek:auto-self; poiesis, creation;)describes vivid systems (GLU) which are characterised by theability to produceand recreate themselves through theelements they are consisted of. This is an essential concept of social system theory.Althoughthisconcept is not newbut its applicationincontext of designing innovative andinteractive web-interfaces hasnot beenexplored. Therefore,this paper does notaim at atechnological solution butgives anew perspective, whichhas implications fordesigning andapplyingtechnology, whichIconsiderfundamental. That view helps to know what thedesign-interventionisaimingat.

137 2.2The possibility of design-interventions in social systems

Thenatureofasocialsystemlike an onlinecommunity is to search foraviable structure. Target control fromoutside of thesocial-systemcannot be reached, as it is autopoietic andtherefore limitedbystrongsystemborders (Luhmann, 1991).Itonlycan self-organise andre-produceitself –through communication. Once an intervention is launched intoasocial system, the designerdoesn’t have influenceonhow it will react. The social system can only understand andinterpret theintervention throughitself. “This avoids theillusionofcontrol (through design) in social systems”(Jonas, 2004). For example, Facebook is asocial system,which constitutes anetworking effect of communication. It is notpossible to force to have many users in it andbuild up an artificial network.

However,whenadesignerdoesnot have control over the system,what is theuse of design?Here Ifollowthe discussion of -Jonas(2006)who considers designing as an evolutionary process of variation, selection andre-stabilisation. Thedesignerdoeshave indirect influence on thecommunitybygenerating variation (newand different interfaces)although she does not have control overthe interpretation andadoption of the system.“Interventioncan only be successful as long as the structural conditions of the specific systemare considered (Jonas2001)”. The system is structurallyconnected with its context(Willke, 2001). Therefore, thecontextual components likethe stakeholders should be involved to createvariations.Consequently, it is necessary to extend the notion of the interface (see Figure 1) to include its context.And the stakeholders of the system is partofthe context.Thisapproachallowsgenerating most appropriate intervention for variationswiththe user themselves. Thedesigner then takesthe role of an agent and mediator between the artefactual system(web interface) andthe context (stakeholders).

Figure 1: Theinterface space(white) as the field of intervention (Jonas 04)

As technical development formsand takes part of theinteraction of an online community: designing variations for web-interfaces becomes agreat challenge. Considering the constitutivecharacteristics of asocial systemand being aware of what thedesign-intervention is aimedat, Isuggest following tools forintervention.

138 2.3Methodologicalapproach

Themethodological approach is qualitative..The focusliesinconcentrating--onone specific,ratherthanonmanyresearch objects, which will be exemplified later in acase study.Ithappens inevitablythatthe researcherregardshim-orherselfasapart of the observation. Duetothe “interpretativeparadigm” (Lamnek,2005)thatsocial reality can onlybeperceived by interpretation.

Thequalitativeapproach serves to perceive and sensitize to aspecific systemby observing:regularities, recurrences, noticeable differences, variations and antagonisms. Irregularities are for instance asuddendifferent salutation of some new members. The way thecommunity system confronts this newcommunicationpattern gives important insights andknowledgeabout thecommunity itself.AsIdo notconsiderthe case study out of ascientific, butout of adesign researcher perspective, thefocus liesingenerating variationsfor interventions. Following the’Jonasian Toolbox’ theinnovation primary starts with projections,where themainquestions are notconcerned with “how it is”, but “howitcouldbe-the ideal”(Chow, 2008). Consequently, Iproposenot only to analyse but also to generate and to filter intervention forvariations with theusers (see also 3.2).

In summary, the qualitative research methods help to open up the interface because they help:

• to considerand be awareofthe specific community segment

• to informthe design process through activeobservationand user participation

• to create and to filter interventions closely between theplatform andthe users

Thequalitative approachenlarges the interface space. It supports to adopt more efficiently designinterventionsbetween the user requirementsand thetechnological platform –through an iterative process of actionand reflection. With thefollowing case, we will demonstrate andexemplify how theconceptofthe autopoiesis is considered in a redesign of an already existing platform of an online-community.

3Case

3.1Design ResearchNetwork -Problem description

Theonline-platform “DesignResearch Network”(seefigure:2)isayoung community, foundedinSeptember 2007forcontributions of Design Research. The platform was created to serve the growing numberofPhD-Students in Design.Objectiveofthe redesign was to increase usability. Until then theplatform wasregarded as rigid and overloadedwithinformation.Besides, more appropriate technological features were believed to raisemorecommunication and therefore activate the users.

139 Figure 2: DRN before the redesign (http://www.designresearchnetwork.org/drn/node), 10.1.08.

140 3.2Interventionstrategy

The strategy was agradual increase in intervention betweenactiveobservationand user participation.Itmustberemindedthatsocialsystems,incontrast to technical systems, are incalculable.They can change withoutpremonition. As it is adynamic system, intervention are always “one shotoperation”(Rittel). It is notpossible to learnout of mistakes. Every trialcounts. If problemsemerge, every oneisindividual. Consequently, aqualtiativeapproach helpstoreconsider theautopoietis of asocial system. Before initiatingirreversible consequences, it was importanttoget to knowthe systemas profoundly as possible, before launching interventions.

Figure 3: Methods used to informthe design intervention

141 The first method was Content Analysis. Theaim was to observe anddescribethe online community withoutinterferingorchanging it. That gave thechancetoget deeper insights of its specific characteristics,without judgingprematurely. After initial impression of thecommunity, Participatory Observationwas conducted.That comprised thechanging positionbetweenscientific observing andparticipatingasauser. Thenext step wasaimed at evaluating the the perspectives of the users. Therefore,questions were askedabout personal experiences with the platform andtheir concrete wishes fornew possible designsand interventions.Figure3showsthe setofmethods appliedduringthe whole design process (see figure 3).

All suggestionswere further developedbyall participating members and technological advances,which Ihaveexaminedinbetween. At the end, all informationwas collected, categorisedand represented in afinalpaper prototypingwith aquestionnaire of possible newfeatures. Finally, themostwanted featureswere illustrated by scenarios.They allowed anticipate possible futuresand give thewished featuresofthe usersa meaningfuland desirable context. These last two methodshelped to project, filter and decide forthe final implementationphase of theproject.

Parallel to thecase study Igenerated knowledgeabouttechnological possibilities, for instance analysing existing features andprinciples of web-experts, successful online communities, as well as searchingfor trends, user behaviour and needs expressed on onlinecommunities. This part of generated knowledgeisnot part of this paper, as it mainly refers to theindividuallearning process duetothe specific community.

3.3 Results and reflections

Theresults of thefinal online-questionnairehaveshown that community members were particularinterested in newinteractivetools.Especially thosethatallowed users to use the interface without confinedtoatoo rigid content navigation.

Anothermajor interest was auser generated calendar relevant to design events. In this calendarspecific eventscan be selected that might be relevant forthe specific community. News aboutdesigneventsthatare in putbythe users.Itwas recommended that the calendar wouldbeaccessible as amashup,besides, it couldbeintegrated with external Blogs, as well as the possibility to receive new information by news-feeds.

Thenextmostpreferred toolwas page views. That supports theconfirmingfeedbackof otherusers.Althoughauserpublishes acontribution andothersdonot comment it, a great numberofpage views show an already existinginterest of thecontent. Thefurther wish, to have more possibilities to connect witheachother,for instancethrough same interests, or research themes, serves the need to build up apersonal social networking system.

142 The last most importantneedwas apersonalisedlogin. While someusers liketohave access to all features, others only needafew important functions.Thisfunction was combined with the principle of social software(O´Reilly, 2005). Everyfunction wasset up like asoftware windowthatcan be reduced, removedand fadedin. This allows adopting theinformationarchitecture to serve theindividualneedsofthe users. Figure 4 shows theweb-platform after redesigning.

Figure 4: Redesign of DRN; with amore flexible interface accordingtoindividualuserneeds & wishes.(http://www.designresearchnetwork.org/drn/node), 10.1.2008.

143 Partly, Iwas surprisedbythe chosenfeatures, for example by the calendarfor design events.Ihad conjecturedthat the users would vote primary for featureswhichincrease usability, likeatoolbartoformattextoraninformationarchitecture which helpsto categorise and visualise information as proposed by others. Theunexpected outcome madeclear that anticipationmight lead to inappropriate conclusion.Consequently, it was importanttoinvolve theusers’expertise.

The whole redesign wasalearning process betweenactiveobservationand auser centred participatory approach. Especially the users helpedtounderstand the specific emergence of thegroup and to inform thesocial centred design. Ideas were constituted and confirmedbydifferent prospectiveand experiences.

4Conclusions

We havelearned that the notion of aweb-interface as astrong cutbetween the users and thecommunity reducesthe community to amerely platform for web-technology(cp. figure 5a). Technological applications, features andprinciplescan only procure with a high technological platform but not with asocial system like an online-community. For thefirst step, it seems usefultochangethe focus on thecommunity andtoput it into the centre (cp.figure 5b). Thelast andfinal step visualises, that dealingwith communities it is useful to considerthe vividdynamicand autopoietic characteristic of thecommunity (cp.figure5c).

Figure 5: DesigninginnovativeWeb 2.0 interfaces leadstodealing with uncertainty of social systems

The extendednotion of the interface made it possible to makeaware of the constitutive characteristics of asocialsystem: the autopoiesis, which can be compared with an individualcollectiveidentity. It is not possible to getdirect feedback outofthe systemor to interfere with it directly; therefore, we hadtotakecare not to interpret anyeffect prematurely. As thecommunity cannot be interfered with interventions directly, generating variations onlyseems to make sense by involvingthe users’expertiseand commitment.

144 After theintensiveexaminationofthe platform DRNetwork,the users andthe community Icametothe conclusion that ahuman andsocialcentred webinterface is set up primary by sufficientstandards that givesenoughstructure andsuggestions how to inform and commitcommunication. Secondly, with an interface that provides enough space andflexibility to adopt to individual needs andexpressions (for example direct feedback) of the users. Nevertheless, the expert attitude of an “optimal web-interface” is not reachable. Web-interfacecan only be appropriate and suitable to aspecific user group.

Consequently, dealing withweb 2.0 interfaces,adesignercan not be regarded as imperator of anew empire, but much more as an agentand mediator among others, interactingand creating designsolutions betweenasteadily changing andevolving socialsystemand its context.But as the future can’tbepredicted, only anticipated by reactions of the community, the designer (andthe solution system) can notbemorethan an perturbator andexperimentator in search for gettingmoreknowledge aboutthe social community.

5References

[Al07] Alby, T.: Web 2.0, Konzepte, Anwendungen,Technologien,CarlHanser Verlag, München, 2007. [Co08] Chow, R.: Case Transfer Vs Case Study: An Evaluation of Case StudyasaMethod for Design Research,Swiss Design NetworkSymposioum 2008 „focused“, Basel Schweiz, 2008. [Jo04] Jonas, W.:Researchthrough design –lectureonthe conference of the swissdesignnetwork,Basel,Schweiz, 13. und 14. Mai. 2004 http://www.conspect.de/jonas/PDF/Jonas_SDN_Basel_05_04_final.pdf [Jo04] Jonas, W.: Mind the gap! -onknowing and not-knowing in design, Hauschild, HM, Bremen2004. [Kr94] Krippendorff, K.:Der verschwundene Bote. Metaphern und Modelle der Kommuni- kation.In: Merten, Klaus/Schmidt, SiegfriedJ./Weischenberg (Hrsg.): Wirklichkeit der Medien. EineEinführungindie Kommunikationswissenschaft, Opladen,S.79-113, 1994. [Lu05]Luhmann,Niklas: Einführung in die Theorieder Gesellschaft, Suhrkamp-Verlag, FrankfurtamMain, 2005. [RiM37] Horst W. J. Rittel, Melvin M. Weber: DilemmasinaGeneral Theory of Planning.Policy Sciences 4(1973), 155-169 ©Elsevier Scientific Publishing Company, Amsterdam -Printed in Scotland [So05] Surowiecki,J./Beckmann, G.:Die Weisheit derVielen, Bertelsmann Verlag, Berlin, 2005. [Wi02] Willke, Helmut: Dystopia,Studienzur Krisis des Wissens in der modernen Gesellschaft, 1. Auflage, Suhrkamp-Taschenbuch Wissenschaft, FrankfurtamMain, 2002. [Wi01] Willke, H.: Systemtheorie. UTB für Wissenschaft,3.Auflage, Stuttgart,2001.

145

Session5

Graph Theory, Routing andLayering

Agame theoretic approach to graph problems

Thomas Bohme¨ and Jens Schreyer Technische Universitat¨ Ilmenau [email protected], [email protected]

Abstract: We investigate some well known graph theoretic problems from agame theoretic point of view. To coloring and matching problems we associate binary payoff games where the players are the vertices of the graph. Solutions to the graph problems correspond to action proﬁles of the game, where all players get payoff1.Weshow, that there exist rules for the choice of action in the repeated play of these games, that convergetothe solution of the graph problems. Although the convergence is slow, this shows, that the problems can be solved with almost no information on the underlying graph.

1Introduction

Manyclassical graph problems, that are well observed for the case, that the underlying graph is part of the input become much more challenging if the graph is an existing network and there is no global instance to solvethe problem. It is the task of the nodes which are the decision makers to solvethe problem, using only information on their neighborhood in the network. Starting with the pioneering work of Linial [Li92] arich literature of what can be done and what cannot be done using only such local information emerged in the field of distributed computing. However, to the best of the author’sknowledge there is no generally accepted definition of local algorithms. Some authors restrict the knowledge of one agent to some small part of the graph near the vertexcorresponding to the agent[An07], others alloweach vertexonly to communicate with its immediate neighbors [LOW08, KMW04]. We propose an approach inspired by the paper of Kearns et al[KSM06] on an experimental study of social network behavior.Their idea is, that the agents corresponding to the nodes of the network graph have no common goal buteach of them has aselfish incentive and the solution of the graph problem corresponds to aNash equilibrium or other suitable solutions of agame that reflects these incentives. In their study the authors investigated the usual vertexcoloring problem and the test subjects corresponding to the vertices of the graph got moneyiftheywhere successful in choosing a color distinct of the colors chosen in the neighborhood in one round of the game. Recently, Chaudhuri et al. [CCJ08] theoretically investigated this game. Theycould show, that if the number of colors is at least Δ+2,then with high probability 1 − δ the graph is colored properly within O(log(n/δ) rounds, where n denotes the numberofvertices. We propose to generalize the idea by investigating for which graph problems it is possible to design such agame, where the players are the vertices of the graph and asolution of the game is asolutionfor the graph problem. Moreover, the payofffunction for each

149 player should depend only on the actions of the neighbors or be at least easily accessible. Computing pure Nash equilibria of games on graphs can be very hard even if centralized computation is possible [DT07, ZCT08]. We are interested in wether there exist adaption rules for the players such that the repeated play converges to asolution of the game. The information used for the adaption process are only the receivedpayoffand the action of a player in one round. This defines anew concept of local computability of graph problems which reflects the possibility of self organization of large networks without global knowledge. In the present paper we investigate some coloring and matching problems fitting to the framework. Our results show, that very simple algorithms convergetooptimal solutions of the problem, butitmay takealot of time.

2Preliminaries

Throughout this paper we consider simple ﬁnite graphs G =(V,E),where V =[n]= {1,..., n}.Aproper k-coloring of G is afunction c : V → [k] such that for all edges ij ∈ E we have c(i) -= c(j).Weidentify such acoloring with the n dimensional vector (c(i))i∈[n].The smallest number k such that G has aproper k-coloring c is called the chromatic number χ(G).

Problem 1(VERTEX COLORING) Givenagraph G and the chromatic number k = χ(G) compute aproper k-coloring c of G

Awell-known extension of this problem is the list coloring problem. Apart from the graph G we are givenalist assignment l : V → 2N,where l(i) ⊆ N denotes the set of admissible colors for the vertex i ∈ V .Aproper l-coloring of G is afunction c : V → N with the property that c(i) ∈ l(i) for all i ∈ V and c(i) -= c(j) for all edges ij ∈ E. G is called l-colorable if there exists aproper l coloring of G.

Problem 2(VERTEX LIST COLORING) Givenagraph G and alist assignment l such that G is l colorable compute aproper l- coloring c of G

If l(i)=[k]for all vertices i ∈ V this is equivalent to problem 1.

A matching of G is asubset M ⊆ E of edges with the property,that no twoedges of M have an end vertexincommon. M is maximal if all edges outside M have an end vertexin common with an edge in M.Amaximum matching is amaximal matching with maximum cardinality and a perfect matching is amatching M where every vertex i ∈ V is an end vertexofanedge in M.

150 Problem 3(PERFECT MATCHING) Givenagraph G containing aperfect matching compute aperfect matching of G.

Note that while deciding whether agiven graph is k-colorable is NP-complete [GJ79], a maximum matching can be found in polynomial time [Ed65].

Aﬁnite game Γ=(N,A, u) consists of

• Aset N = {1,..., n} of players. • Forevery player i ∈ N aset Ai of actions and A = A1 × ... × An.

i i • Apayofffunction u =(u)i∈[n]where u : A → R denotes the payofffunction of player i.

An element ai ∈ Ai is called action and an element a =(a1,..., an) ∈ A is an action profile.The payofffunction associates to every possible action profile apayofffor every player.Let S be asubset of players. By (a−S,bS)we denote the action profile, where each player i ∈ S chooses action bi ∈ Ai and all players i/∈Schoose ai ∈ Ai.A(pure) Nash equilibrium of Γ is an action profile a ∈ A where no player has an intention to deviate, i.e.

∀i ∈ N∀bi ∈ Ai : ui(a) ≥ ui(a−i,bi)

The games we investigate in this papers have the property that there are only twopossible payoffs 0and 1. We call such games binary payoffgames.Weinterpret apayoffof1asa win and apayoffof0asaloss. An action ai ∈ Ai that ensures apayoff1for player i independent of the other player’sactions is called a winning strategy for player i.Anaction profile aS for asubset S of players that ensures apayoffof1for all players of S regardless of the actions of N \ S is called cooperative winning strategy (cws) for the players of S.A subset S of players having acooperative winning strategy is called potentially successful. If the game Γ is repeated infinitely often, the players may adapt there choice of action. Depending on the action of aplayer and the receivedpayoffinone round of the game Γ a probabilistic 1-recall learning rule computes aprobability distribution on the set of actions according to which the action for the next round is chosen. If every player uses such alearning rule, this induces aMarkov chain on the set of action profiles.

We associate binary payoffgames corresponding to the three graph problems stated above. In every case, the players of the game are the vertices of the graph.

COLORING GAME Γ1 =(N,A, u)

• N = V • Ai =[k] • ui(a)=1 ⇔ ai-=ajfor all ij ∈ E

151 That means, every player chooses acolor and gets payoff1ifher color is different from the colors of the neighbors. This is the game ﬁrst introduced by Kearns et al. in [KSM06]. Aproper k-coloring of the graph corresponds to aNash equilibrium of the game, since every player gets maximum payoff. If k ≥ Δ+1where Δ is the maximum degree of G the opposite is also true. To see this, consider aplayer who receivespayoff0.The neighbors use at most Δ different colors, so the player can choose at least one different color and receive apayoff1.This shows, that in aNash equilibrium a every player must get payoff1,which means that a must be aproper coloring. In [CCJ08] the authors propose aprobabilistic 1-recall learning rule, such that the corresponding Markov chain converges to aproper coloring, giventhat k ≥ Δ+2. If k ≤ Δ Nash equilibria do not coincide with proper colorings in general. If G is for instance the complete graph on V =[n]with the edge connecting the vertices 1and 2 missing, the following coloring is aNash equilibrium: ai = i for 1 ≤ i ≤ n − 1 and an = n − 1.All players apart from n − 1 and n receive payoff1,and the last twovertices cannot increase their payoffsince all colors appear in their neighborhoods. Nevertheless the graph is (n − 1)-colorable. If the graph G is the complete bipartite graph Kr,r the situation is even worse. There are twoproper 2-colorings. But almost every coloring is aNash equilibrium. As long as both colors appear in both partite sets, every player gets payoff0 and cannot increase the payoff, since both colors appear in the neighborhood. That means there are 22r − 4 · 2r +6colorings only 2ofwhich correspond to proper colorings. On the other hand, an action proﬁle aS for asubset S of players is acws if and only if all pairs of adjacent vertices in S are colored differently and all players outside S are in different components than the vertices of S.Otherwise, avertexoutside S adjacent oa vertex i ∈ S could choose color ai and player i looses. That means in case G is connected, the only potentially successful set of players is the set of all vertices.

The game associated with problem 2isthe following:

LIST COLORING GAME Γ2 =(N,A, u)

• N = V • Ai = l(i) • ui(a)=1 ⇔ ai-=ajfor all ij ∈ E

Again, every l-coloring of G corresponds to aNash equilibrium of the game, butthe opposite is not true in general. Asubset S of players is potentially successful with acws aS if every pair (i,j)of adjacent vertices of S is colored differently by aS and for every edge ij with i ∈ S and j/∈Swe have ai ∈/ l(j).Weclaim that every such partial coloring aS can be extended to aproper l-coloring of the whole graph G. If b ∈ A is anyaction proﬁle corresponding to aproper l-coloring of G,then (b−S,aS) is also aproper l-coloring. This is the case because by deﬁnition bi -= bj for all edges ij with i, j/∈S,ai-=ajfor all edges ij with i, j ∈ S and for all edges ij with i ∈ S and

152 j/∈Swe have ai -= bj because ai ∈/ l(j).

Forthe perfect matching problem we consider twodifferent games.

FIRST MATCHING GAME Γ3 =(N,A, u)

• N = V • Ai = N(i)={j∈V |ij ∈ E} • ui(a)=1 ⇔∃j∈N(i):ai=j∧aj=i

Foranaction profile a we consider the set M(a)={ij ∈ E | ai = j ∧aj = i}.Obviously M(a) is amatching for all action profiles a ∈ A.Anaction profile is aNash equilibrium of Γ3 if and only if M is amaximal matching, and aset S is potentially successfull, if and only if there exists amatching M such that S is the set of all end vertices of M.

SECOND MATCHING GAME Γ4 =(N,A, u)

• N = V • Ai = N(i)={j∈V |ij ∈ E} • ui(a)=1 ⇔∃j∈N(i):ai=j∧aj=iand for k -= jak-=i

The set M(a) for an action proﬁle is deﬁned as above.Again a is aNash equilibrium if and only if M(a) is amaximal matching. However, the only potentially successful sets of players are all vertices of one component of G or all vertices of the union of some components of G,given that G has aperfect matching.

3Results

We propose the following simple trial and error learning rule for aplayer i the game Γ2.

1. Choose equiprobably arandom color out of l(i) for the action ai in the ﬁrst round of the game. 2. After round t (t ≥ 1) keep the current color if round t is won, otherwise choose equiprobably arandom color out of l(i) for the next round.

i Nowconsider the Markov chain (Xt)t∈N on the state space A where Xt denotes the color chosen by i in round t,ifall players apply the trial and error rule.

Theorem 1 If all playersinthe game Γ2 act according to the trial and error rule,with probability 1thereisatime T after which Xt = c for all t ≥ T ,where c is aproper l-coloring of the graph G.

153 Proof. Forevery proper l-coloring c the set Sc = {c}⊆Ais an absorbing subset of the state space of the Markov chain (Xt)t∈N,i.e. once the trajectory of the Markov chain hits the set, it will neverleave it. This is obvious, since every player gets payoff1ifall players choose acolor according to c.Since the state space is finite, all we have to show is, that these are the only minimal absorbing subsets of the space state. Assume there is asubset S ⊆ A which is minimal absorbing, i.e. anytrajectory of the Markov chain hitting S cannot leave S and S contains no proper subset which is absorbing. Thus, no action profile in S corresponds to aproper l-coloring. Let c be aproper l-coloring and a an action profile in S where amaximum number of vertices chooses acolor according to c.Since a is not aproper coloring, there must be an edge ij such that ai = aj.But then at least one of the twovertices, say i is not colored according to c.Since both players loose in a with positive probability i changes the color to ci and all other players keep their colors. That means (a−i,ci) is in S which contradicts the choice of a.Hence, the only minimal absorbing subsets of A are singeltons corresponding to proper l-colorings of G which provesthe statement of the theorem. ✷

Since the game Γ1 is aspecial case of Γ2 the theorem also applies to usual vertexcolorings. Forthe game Γ4 we consider the the following trial and error rule for aplayer i:

1. Choose equiprobably arandom neighbor out of N(i) for the action ai in the ﬁrst round of the game. 2. After round tt≥1keep the current choice if round t is won, otherwise choose equiprobably arandom neighbor out of N(i) for the next round.

i We consider the Markov chain (Xt)t∈N on the state space A where Xt denotes the neighbor chosen by i in round t,ifall players apply the trial and error rule

Theorem 2 If all playersinthe game Γ4 on agraph G having aperfect matching M act according to the trial and error rule,with probability 1thereisatime T after which i j Xt = b for all t ≥ T ,wherethe set M(b)={ij ∈ E | b = j ∧ b = i} is aperfect matching of the graph G.

Proof. Forevery action profile b where M(b) is aperfect matching the set {b} is absorbing, because all players get payoff1and will not change their actions. We argue that these singletons are the only minimal absorbing subsets of A.Suppose there is adifferent minimal absorbing subset S,then S cannot contain an action profile b where M(b) is aperfect matching. Foraperfect matching M let b ∈ A be the action profile with bi = j ⇔ ij ∈ M Let a be an element of S with amaximum number of players i with ai = bi.Since M(a) is no perfect matching there is an edge ij such that ai = j and aj -= i.Thus both players get payoff0inaand may change their actions in the next round. Foratleast one of both players, say iai -=bi.But then with positive probability the action profile in the next round is (a−i,bi) ∈ S which contradicts the choice of S. Thus, the only minimal absorbing subsets of the state space are singletons corresponding to perfect matchings, which provesthe statement of the theorem. ✷

154 Remark If all players in the game Γ4 on aconnected graph G without aperfect matching M act according to the trial and error rule, with probability 1the play of no player will converge to aconstant play.

Theorem 3 If all playersinthe game Γ3 on agraph G without isolated vertices act according to the trial and error rule,with probability 1thereisasubset S of playersand a S S i j time T after which Xt = b for all t ≥ T ,wherethe set M(b)={ij ∈ E | b = j ∧ b = i} is amaximal matching of the graph G and S is the set of all end vertices of M(b).

The proof of Theorem 3issimilar to that of Theorem 2and is omitted here.

4Concluding Remarks

The results of the paper showthat graph problems can be solved using learning algorithms in suitable binary payoffgames. Morover, the information on the structure of the graph, needed by the players of the game is very little. In fact, theyneed nothing more than the ownpayoffand therefore do not even have to be able to observethe actions of the neighbors directly.Onthe other hand, convergence needs alot of time. But since the coloring problems are NP-complete, efﬁcient algorithms were not to be expected. The results seem to indicate, that the simple trial and error rules lead to convergence to a desirable outcomes likeNash equilibria or cooperative winning strategies in anybinary payoffgame. That this is not the case wasshown by an example of Hart and Mas-Colell (Theorem 1in[HM06]).

References

[An07] Andersen, R.; Borgs, C.; Chayes, J.T.; Hopcroft, J.; Mirrokni, V.S; and Teng, S.: Local Computation of PageRank Contributions. WAW2007; pp. 150-165.

[CCJ08] Chaudhuri, K.; Chung, F.;Jamall M.S.: Anetwork coloring game. WINE 2008; pp. 522-530.

[DT07] Durr¨ ,C;Thang, N.K.: Nash Equilibria in Voronoi Games on Graphs. ESA 2007; pp. 17-28.

[Ed65] Edmonds, J.: Paths trees and ﬂowers. Canadian journal of mathematics 17 (1965); pp. 449-467.

[GJ79] Garey, M.R.; Johnson D.S.: Computers and intractability: aguide to the theory of NP- completeness. W.H. Freeman, NewYork, 1979.

[HM06] Hart, S.; Mas-Colell, A.: Stochastic uncoupled dynamics and Nash equilibrium. Games and economic behavior 57(2) (2006); pp. 286-303.

155 [KSM06] Kearns, M.; Suri, S.; Montfort, N.: An experimental study of the coloring problem on human subject networks. Science Vol313 (2006); pp. 824-827.

[KMW04] Kuhn,F.; Moscibroda, T.;Wattenhofer,R.: What Cannot Be Computed Locally! 23rd ACMSymposium on the Principles of Distributed Computing (PODC), St. John’s, Newfoundland, Canada, July 2004.

[LOW08] Lenzen, C.; Oswald O. A.; Wattenhofer,R.: What can be approximated locally? 20th ACMSymposium on Parallelism in Algorithms and Architecture (SPAA), Munich, Ger- many, June 2008.

[Li92] Linial, N.: Locality in distributed graph algorithms. SIAM J. Comput 21(1) (1992); pp. 193-201.

[ZCT08] Zhao, Y.;Chen, W.;Teng S.: The Isolation Game: AGame of Distances. ISAAC2008; pp. 148-158.

156 Splitting Overlay Networkfor Peer-to-Peer-based Massively MultiplayerOnlineGames

ChengLIU,Wentong CAI

Paralleland DistributedComputingCenter School of ComputerEngineering Nanyang Technological University Singapore639798 {LIUC0012, aswtcai}@ntu.edu.sg

Abstract: Massively MultiplayerOnlineGames (MMOGs)gradually become one of themostpopular internet applications.Traditionalclient-serverarchitectureiswidely used in MMOGs’deployment, butits scalability andmaintenance arelimitedmostly by servers. Peer-to-Peer (P2P)architecture, whichattempts to makeuse of computer resources fromthe computersinthe network, is considered to be acandidate infrastructure forMMOGs.Inthispaper,weinvestigatesomerelated work andpropose an algorithmtosplit overlay networkfor P2P-basedMMOGs.Weshowthe beneﬁt of splittingthe overlay networkinreducinglookup latencyofgame objectsthrough experiments andanalytical analysis.

1Introduction

Massively MultiplayerOnlineGames (MMOGs)are growingrapidly with thedevelopment of computingpower andnetwork.Theyprovide gaming environment forhundreds of thousands of playersaround theworld.AnumberofMMOGs,suchas[wow04],have great commercialsuccessand attract more andmoreattentioninrelated research ﬁelds. Client-serverarchitectureiswidelyemployed in thecurrent MMOGimplementations. Players can accessthe game worldbyconnectingtothe centralized serversusing their owncomputers. Themoreplayers one server holds,the morerevenue it maygeneratefor game vendorswho maintain theservers.However,the connected playerswill consume the bandwidth andcomputingpower in theserver. Therefore, clusters of serversare usually used to manage game play to make MMOGs scalable.But theincreaseinthe numberof playersmay require more serverstobeaddedintothe clusters. Peer-to-Peer (P2P)architectureisanewdistributed computerarchitecturedesignedfor sharingcomputer resources.Ithas been employed in differenttypesofapplications,such as instantmessaging applications,distributed computingand ﬁle sharing. Without the necessity of acentralized server or an authority,resources distributedinthisarchitecture can be accessedbyjoinedpeers. Because it is designedtobeself-organized andhas good characteristicsinscalability androbustness, peer-to-peer architectureisemergingtobea

157 suitablearchitecturefor distributedapplications,including MMOGs. Thereare hundredsofthousands of items or objectsinthe virtualworld of MMOGs.Those objectscan be accessedsimultaneously by players. Twomajor problemsinMMOGare howtoinformaplayer theobjectsorplayers nearby in thevirtual environment andhow to keep game objects’ states consistent among players. In client-serverarchitecture, servers areusually employed to collect thestatesofall game objectsand inform theplayers the updatedgamestates. However, in aP2P infrastructure, players’ computers serveaspeers in theoverlay networkand thegameobjectsare created andmaintainedbythese peers. During game-play, theplayers need to raisequeries ﬁrst to obtainthe game states near its positioninthe virtualworld.The experience of theplayers can be improvedwith less networklatency in game objects’ lookup. Therefore, themainobjectiveofthispaper is to reducethe lookup latencyinaP2Poverlay networkfor MMOGs. We will presentthe existingrelated work in Section2and proposeasplit algorithmin Section3.After that,wewill demonstratethe beneﬁt of reducinglookup latencythrough bothexperimentsand analytical analysis in Section4.Wewill discussmoreissues in Section5and conclude thepaper in Section6.

2Related Work

P2Poverlay networkstructurescan be categorized into unstructured and structured ones accordingtotheir contentplacement. In unstructuredpeer-to-peer overlay network, content can be put anywhere andthe queries can be implementedthrough some mechanisms(e.g.,flooding);whereas,the overlay is controlleddelicatelyinstructurednetwork wherecontents areplaced at some specifiedlocations andthe queries can be routed efficiently through distributedroutingtables. Acommon idea of usingP2P networkinonlinegames is to make thewholebig game worldpartitionedintomultiple regions whichare then assignedtopeers. B. Knutsonetal. proposed theirMMOGarchitecturenamed SimMud[KLXH04] whichwas implemented on ageneral structured P2Poverlay namedPastry[RD01]and ascalable applicationlevel multicastinfrastructure(i.e.,Scribe[RKCD01]). Gameworld is dividedintoseveral fixed rectangleregions,where playersinthe same region communicatewith each otherina multicastgroup managedbyScribe. Allpeersand thegameregions aremappedtouni- formlydistributed IDsina128-bits name space in Pastry.The peer,who hasthe closest ID to agameregionID, is chosen to be coordinatorfor that game region. Coordinators are not onlyresponsible formessage gatheringand synchronization, whichare relatedtothe events happenedinthe game region, butalsoact as theroot of amulticasttreefor message delivery.Players in differentregions mustcommunicatewith thehelpofthe coordinators. As it supports a d-dimensionalvirtual space fordatalocation, thestructuredP2P network CAN[RFH+01]providesastraightforwardway to mapthe partitionedgameregiononto servers[RWF+ 07]orsupernodes[RMO08] whichare selected frompeers. Gamestates aremanaged by theserverorsupernode accordingtotheir locations in thegameworld, andevery server or supernode onlyneedstoknowthe 2d direct neighbors. With thefunc-

158 tionality (e.g., message routing, topology updating) of CAN, theregions can be merged or split dynamically.Moreover, Thereplicationofthe game states in neighborscan help in node failure recovery as wellasimproving of lookup performancewhenthe player needs theinformationfromthe adjacentgameregions. Colyseus [BPS06] is adistributed architecturefor multiplayergames andamodifiedQuake II is supported. It wasimplementedonarange-queriable structured overlay,calledMer- cury [BRS02].Mercury creates a routehub foreach attribute(e.g.,different dimensions) in theapplicationschema. Meanwhile,itorganizes peersinacircular overlay while keep- ingadjacentpeersresponsible foracontiguous range of keys.Instead of region-based partitioning of game world, area of interest (AOI)filtering is implementeddirectly in this range-queriable overlay.Moreover, object locationmetadataand queries arelikelytoex- hibitspatiallocality,and thus can be mappeddirectly ontothe overlay.Thisallows the playerstocircumventroutingpaths andget theneeded objectsbycachingrecentroutes. In thetraditionalDHT (Distributed HashTable)protocols such as Chord[SMLN+03], CAN[RFH+01]and Pastry [RD01],each peer in thenetwork is assignedaunique identi- fierand is responsible foracertain part of keyspace equally.The query of keys arerouted closer to thepeer whoseidentifiermostclosely matchesthe key. Some hops mayincur largenetwork delaybecause peersmay routethe messagestoafar-located peer in the underlying network. In ordertoovercome this problem,the topology aware lookup protocols (e.g., [RGRK04])wereproposed by consideringthe proximity of peers. However, theextra storageand communications arerequiredtocreatethe secondary lookup overlay with peersthatare located closelyaccordingtothe physical topology. Some researchersfocused on reducingthe numberoflookup hops through alarge indexof peers. Li et al proposed aDHT protocol calledAccordion[LSMK05]inwhich therouting tablesizecan be adjusted accordingtothe rate of churnand networksize. It can achieve O(1) lookup latencywhenbandwidth is plentiful andchurnislow,and O(logN) lookup latencyinhighchurnenvironment andthe availablebandwidth is low. Parallellookup [LSM+05]and replicationoflookup key[DLS+04]are twocommon but important methods to improve lookup performancefor DHTs, especially underchurn. In a parallellookup, multiple lookups areinitialized simultaneously by theoriginator. Together with iterative routing, multiple copies of query messagesare sent out in each hops of parallellookup. Thewholelookup processcan continue without beingblocked even when some stalepeersare met, so lookup retrycan be avoided. In additiontoimprove thelookup performance, keyreplicationcan also handlethe problemscausedbychurn. By copying thedatakeystoother peers, thelookup can still getthe result even though thepeer whichis responsible forthe data keyleavesthe system.Mostofthe current research is concentrated on themethods to choosethe suitablepeerstoput areplicaofthe data key.

3Our Approach

We followthe approach of DHT in MMOGs.Weobserve that players’ cooperationis very popular in modernmassively multiplayeronlineroleplaying games(e.g. ateam of

159 playersfight together to kill amonsterinthe game world).Therefore,inour approach, we concentrateonplayers’actions with game objectsratherthanobtaining information of asmall area in thegameworld,and this motivates us to usethe DHT to query game objectsdirectly rather than game regions.Asthese game objectscan be designedinthe stageofgamedeveloping, we make an assumptionthateach object hasaunique identifier andplayerare aware of thecorresponding identifiers of certain objectswhich it wantsto get. When playingthe onlinegames,players must knowwho arecurrently modifying thestate of theobject they areinterestedinand wheretheycan getthe object’s current state. Thepeerswhich can providethe informationabout objectsare namedsuppliers in our approach.One important issueinonlinegames is that theobject’s stateshouldbe maintained consistantly among agroup of players. This is guaranteed by thegameobject’s supplier. Thesupplierreceivesthe modificationfromthe playersand then disseminatesthe latest game object’s states to theplayers. Rather than discussing thegameobjectsmanagementinDHT,wefocus on aDHT split algorithmtoimprove thelookup of game objects. Through DHT split, theoriginalDHT will be dividedintoseveral DHTs andthe averagenumberoflookup hops can be reduced as thequery messageswill be routed within agroup of peers. Moreover, thenetwork latencyofgameobject’s lookup can be improvedfurther if thegeographical locations of peersare takenintoaccount in determining peer groups.

4Split ChordRing

TheDHT basedonChord[SMLN+03]isemployed in severalP2P applications,sowe take Chordringunderlow churnasthe exampletodemonstrateour DHT split algorithm in this paper. Each peer in Chordisassignedaunique identifierusing hash function. All peershaveone direct successorand predecessor. They form an identifiercircle, named Chordring. Each peer hasakeyspace ranging fromits predecessor’sidentifiertoits own. Thepeer whosekey space covers thehashvalue of keyshouldrespond to thequery for such akey.

%% %#

%$ !*AO ,

Figure1:Chordrings

160 Table1:Definitionofvariables fornode n, Table2:Definitionofneighborsand split usingm-bit identifiers levelinnode n Notation Definition Notation Definition finger[k] first node on circle that suc- neighbor[i] theneighbor in i split ceeds (n +2k−1)mod 2m, 1 ≤ level current split level; k ≤ m times of split successor thenextnode on theidentifier circle; finger[1].node predecessor theprevious node on theidenti- fiercircle neighbor one node in theother ring

As it is showninFigure1,the circle represents theoriginalChordringand theresttwo lines representthe connectionoftwo newrings aftersplit. Peer n2 whosedirect successor in theoriginalringisn3 will take n5 as its newsuccessorafter ring split. Theidentiﬁerof Key1is coveredbyn6’s keyspace in theoriginalChordring. Afterringsplit, n6 and n7 can respond to thequery of Key1.

4.1Single Split

TheoriginalChordringcan be split into twosmall rings.Inorder to avoidtwo isolated rings,peer in thenewly split Chordringwill take one peer in theother ring as its neighbor. The neighbor foreach peer will be decidedduringthe ring split. We followthe notations in Chordand addone neighbor foreach peer in Table1. Algorithm1presents theprocedureofsplittingChordring. It begins with theprocedure to determine whether one peer andits successorwill belong to differentChordrings aftersplit. Checkingisdone accordingtosome generalproperties, e.g.,the geographical locations of peers. The neighbor and successor in each peer mustbeupdatedduringring split. If peer n finds that its successorwill belong to adifferent ring, it will become its successor’s neighbor andask its successortofind anew successorfor it. If n.successor and n.successor’s successor areinthe same ring,theywill become peersinthe newring (bothofthemhavethe same neighbor n). Otherwise, n.successor returnsits successoras thenew successorofn.The fingertable entrieswill be updatedperiodically usingthe original procedureinChord.

4.2MultipleSplitand Forwarding Messages

TheoriginalChordringcan be split forseveral times.One ring will be split into twoeach time. The split level of aringisusedtorepresent howmanytimes thesplit operationhas been performedwhenthe ring is generatedfromthe original one.Two newChordrings will have thesame split level aftereach ring split. TheoriginalChordringisinsplit level 0,and thelevel will be increased by one fromthe formerringafter splitting. Each peer

161 will nowmaintainalistofneighbors, in additiontokeep track of its level(SeeTable 2). Aftereach ring splitting, peersinthe ring will increaseits split level first andadd one more neighbor to theneighbor list. To keep track of thesplit leveland to addnew neighborsto theneighbor list, the send message procedureinAlgorithm1is modified(seeAlgorithm 2). With neighborsindifferent levels,wecan forwardamessage in differentrings wisely and avoidflooding theP2P network. Thereisacorresponding levelfor each message.When amessage is generated, it is sent to everyneighbor in theneighbor list. Thelevel of the message is setaccordingtothe levelofits neighbor in thelist. We describe themessage forwarding in Algorithm3.

Algorithm 1 Pseudocode of splittingaChordring Procedure: n.split() inSameR = check in same ring(n, successor) if ! inSameR then Procedure: n.remove predecessor() successor.remove predecessor() if predecessor is notnil then successor.send message(NEIGHBOR,n) predecessor = nil endif endif Procedure: n.send message(msg id,n!) Procedure: n.fix fingers() inSameR = check in same ring(n, successor) next = next +1 if msg id is NEIGHBOR then { neighbor = n! if next >mthen next =1 if inSameR then endif successor.send message(NEIGHBOR,n!) finger[next]=find successor(n+ else 2next−1) n!.send message(SUCCESSOR,successor) endif } Procedure: n.find successor(id) else if msg id is SUCCESSOR then if id ∈ (n, successor] then successor = n!; n!.notify(n) return successor endif else ! Procedure: n.notify(n ) return ! if predecessor is nilorn ∈(predecessor,n)then successor.find successor(id) ! predecessor = n endif endif

Algorithm 2 Pseudocode of adding newneighbor Procedure: n.send message(msg id,n!) inSameR = check in same ring(n, successor); level = level +1 if msg id is NEIGHBOR then { neighbor[level]=n! if inSameR then successor.send message(NEIGHBOR,n!) else n!.send message(SUCCESSOR,successor) endif } else if msg id is SUCCESSOR then successor = n!; n!.notify(n) endif

162 Algorithm 3 Pseudocode of forwarding messagesinrings Procedure: n.send to neighbors(message) for i ≤ level do if message.level

5Analysis andExperiments

In this section, we evaluate thenetwork latencyoflookup afterwesplit aChordring accordingtothe peers’ geographical locations.Whether twopeersbelong to thesame ring or not can be decidedbyusing landmarksapproach [ZZZ+04]. We assume that the function check in same ring in Algorithms 1and 2isdeﬁnedand thenetwork latencyof thepeersinthe same group is less than thelatency of peersindifferent peer groups.A key-valuepairwhich is held by one peer in theoriginalChordringcan be copied to the peersinthe otherrings aftersplit accordingtothe key’spopularity in differentpeer groups. Hereweconsider twosimplepolices: • Withoutreplication: Afterringsplit, thekey-value pair is still in thesamepeer as in theoriginalring. • With replication: Thereisone copy of thekey-value pair in apeer foreach ring. Therefore, peerscan getthe query result fromapeer in thesamegroup.

5.1PerformanceAnalysis

We ﬁrst consider thedifferenceoftotal numberofhops afterand before ring split without themechanismofreplication. Supposethataring (r)issplit into tworings r1 and r2 with m1 and m2 peersrespectively. Forthe queries raised by peersinring r,the totalnumber of hops changed, δ1,can be calculatedby: m2 log m m2 log m H = 1 log m + m ∗ m ∗ (1 + 2 2 )+ 2 log m + m ∗ m ∗ (1 + 2 1 ) (1) after 2 2 1 1 2 2 2 2 2 2 1 2 (m + m )2 H = 1 2 log (m1+m2) (2) before 2 2

δ1 = Hafter − Hbefore (3)

Hafter represents thetotal numberofhops afterthe ring r is split. After r is split into two, apeer in either ring r1 or r2 can startkey lookup. Thedestinationpeer whichresponds to thequery couldbelong to anyone of thesetwo rings.Four terms in equation(1) represent four differentcombinations of originator-destinationpairs.Ifthe originator anddestination peersare in thesamering(e.g.,ring r1), thenumberofhops is (log2m1)/2 accordingto theresultin[SMLN+03]. But, if they belong to differentrings (e.g., theoriginatorinring r1 andthe destinationinring r2), thenumberofhops in each lookup shouldinclude one hop through theneighbor linkand thenumberofhops to routethe query in theother ring.

163 Assume that thereare n peersthatare not in ring r.Iftheyraise akey lookup forthe keylocated in apeer in ring r,the difference in thetotal numberofhops, δ2,iscalculated as follows.There aretwo possibilitiestoget thekey forthe peer outside ring r –ifits neighbor andthe destinationpeer areinthe same ring (e.g., ring r1), only (log2 m1)/2 hops is needed;elseone extrahop is includedtoforward thequery to theringwhere thedestinationpeer belongs (i.e., ring r2 in this case) in additiontothe numberofhops requiredtoroutethe query in ring r2.Therefore,the totalnumberofhops can be calculated by: m1 log2 m1 m2 log2 m1 n ∗ m1 ∗ ( ) ∗ ( )+n∗m1∗( )∗(1 + ) m1 + m2 2 m1+m2 2 This can be simpliﬁedas:

m2 log2 m1 n ∗ m1 ∗ ( ∗ 1) + n ∗ m1 ∗ m1 + m2 2

Thesamereasoning can also be appliedwhenthe neighbor is in ring r2.Hence, we have:

! m2 n ∗ m1 m1 n ∗ m2 Hafter = n ∗ m1 ∗ ( ∗ 1) + log2 m1 + n ∗ m2 ∗ ( ∗ 1) + log2 m2 (4) m1 + m2 2 m1 + m2 2 n ∗ (m + m ) H! = 1 2 log (m + m ) (5) before 2 2 1 2

! ! δ2 = Hafter − Hbefore (6) So,fromequations (3)and (6), thetotal numberofhops changedbyring r’s split is:

δH = δ1 + δ2 (7) When apeer in the n peersthatare not in ring r triestoﬁnd akey located in one of thesepeers, thenumberofhops of thesequeries will not be affected by theringsplit. The numberofhops of queries is also not considered in theabove analysis when apeer in r triestoﬁnd akey located in apeer that arenot in ring r. Then,weanalyze theimprovement of networklatency forkey lookups afterringsplit. Assuming that L1 and L2 standfor theaverage networklatency foreach pair of peersin rings r1 and r2 respectively, we canget theaverage networklatency of r as

m2 ∗ L + m2 ∗ L +2∗m ∗m ∗L L = 1 1 2 2 1 2 o (8) avg 2 (m1+m2)

Where Lo represents theaverage networklatency of theneighbor links. Following thesimilaranalysisabove,weadd thenetwork latencytoequations (1), (2), (4) and(5).So, thechange of lookup latency, δL ,can be calculatedasfollows:

m2L L log m m2L L log m L = 1 1 log m + m ∗ m ∗ (L + 2 2 2 )+ 2 2 log m + m ∗ m ∗ (L + 1 2 1 ) after 2 2 1 1 2 o 2 2 2 2 2 1 o 2

(m + m )2 ∗ L L = 1 2 avg log (m1+m2) before 2 2

! m2 ∗ Lo n∗m1∗L1 m1 ∗ Lo n∗m2∗L2 Lafter = n ∗ m1 ∗ ( )+ log2 m1 + n ∗ m2 ∗ ( )+ log2 m2 m1 + m2 2 m1 + m2 2

164 n ∗ (m + m ) ∗ L L! = 1 2 avg log (m + m ) before 2 2 1 2

! ! δL = Lafter − Lbefore + Lafter − Lbefore (9)

We also analyzethe numberofhops andlookup delaychangedafter ring split when repli- cationmechanismisused. Since keys arereplicated,inthiscaseaquery can always be answered by apeer in thesamering. In thefollowing analysis,onlyqueries raised from peersinring r areconsidered.Whenapeer in aringraisesakeylookup, thecorresponding averagenumberofhops is (log2mi)/2 where mi is thenumberofpeersinthe newChord ring. Thedifferenceofthe totalnumberoflookup hops afterand before ring r’s split, δH , can be calculatedusing equation(10),where m1 and m2 arethe numberofpeersinrings r1 and r2 respectivelyafter split and m is thenumberofpeersinthe original ring (i.e., m = m1 + m2).

m ∗ m m ∗ m (m + m ) ∗ m δ = 1 log m + 2 log m − 1 2 log (m + m ) (10) H 2 2 1 2 2 2 2 2 1 2 We followthe deﬁnitionofthe averagenetwork latencyinequation(8) andget thechange of thelookup latencyafter andbeforeringsplit as follows:

m ∗ m ∗ L m ∗ m ∗ L (m + m ) ∗ m ∗ L δ = 1 1 log m + 2 2 log m − 1 2 avg log (m + m ) (11) L 2 2 1 2 2 2 2 2 1 2

5.2Simulation

We ﬁrst build asmall game worldwith 200 game objects. Theunderlying networktopology with 300 nodesisgenerated by BRITE [MLMB01]using Waxmanmodel. Thesizeof main planeisset to 1000 andthe propagationdelay between twodirect-connected nodes is less than 5timeunits.The shortest routebetween anytwo nodesiscalculatedusing Dijkstra’s algorithm. 300players aremappedrandomlytothe nodesinthe networktopology.All playersmove randomlyand perform 600 times of game object lookups in the following scenarios: Scenario 0: theoriginalChordringwith 300 peersbeforesplit Scenario 1: twoChordrings with 50 and250 peersafter split fromthe original one Scenario 2: threerings with 50, 104 and146 peersafter theringwith 250 peersissplit Thenetwork latencyfor object lookup is collected andthe distributionisshown in Fig- ure2.68% lookup arelessthan35timeunits in scenario 1, and74% lookup arelessthan 35 timeunits in scenario 2without object replication. Butthere areonly52% of them before split. With theobject replication, 85% and97.5% lookup whichare less than 35 timeunits can be identiﬁedinscenario 1and scenario 2respectively. We useBRITE againtogeneratealarger topology usingthe same setting. 9947 players aremappedrandomlytothe nodesinthe networktopology andthese nodesformaChord ring. Then theoriginalChordringisgradually split into 2, 4, 7, 8and 10 rings andthe peersineach ring aredeterminedbytheir geographical locations.After each split, we move those9947 playersrandomlytoquery an object out of 15,000 objectsinthe game world. Theaverage numberofhops (Hi)and thenetwork latency(Li)for 12000 times

165 1 1 Scenario 0 Scenario 0 Scenario 1+without replication Scenario 1+with replication Scenario 2+without replication Scenario 2+with replication

0.8 0.8

0.6 0.6 CDF CDF

0.4 0.4

0.2 0.2

0 0 0 20 40 60 80 100 0 20 40 60 80 100 Lookup latency Lookup latency (a) (b) Figure2:The distributionoflookup latencywithout/with replication of game object lookups arerecorded.Inorder to check whetherthe lookup performance is improvedwhenthe peer geographical locationisconsidered,wenormalizethe results with thefollowing functions where Horiginal and Loriginal standfor theaverage number of hops andlookup latencyinthe original Chordringrespectively:

E hop = Hi/Horiginal E latency = Li/Loriginal We also evaluate thenumberofhops andlookup latencyaccordingtoour analysis,and compare theresulttothe experiments. Thenumberofhops andlookup latencyafter split- tingseveral times comparing to thenumbers in theoriginalChordringiscalculatedby (12),where δH and δL aredeﬁnedinequations (7)&(9)and (10) &(11).Inorder to simplifythe computation, we supposethat L is theaverage latencybetween peersinthe same ring (soboth L1 and L2 areequaltoL), and Lo = c ∗ L (c>1.0) is theaverage latencybetween neighborsinthe differentrings.There are m peersinthe original ring with theaverage networkdelay among peersisLchord.

Σδ Σδ F hop =1+ H F latency =1+ L (12) 2 2 (m ∗log2 m)/2 (m ∗Lchord ∗ log2 m)/2

Thecomparisonofour experimental result andanalytical result is showninFigure3with c =1.015.Asshown in theﬁgure, theexperimental andanalytical results areveryclose to each other. Theaverage numberofhops andthe lookup latencyisreduced gradually.So we canuse theanalytical formulastoanalyze theperformanceoflarge P2Pnetworksorto analyzeperformanceundervarious networkconditions.For example, theresultofthe network latencyimprovement with differentvaluesofcis showninFigure4.Itdemonstrates that thebiggerthe difference of thenetwork latencybetween andwithin peer groups,the moresigniﬁcantimprovement theringsplittingcan achieve.

166 1.005 E_hop 1 E_hop F_hop F_hop 1 E_latency E_latency F_latency F_latency 0.95 0.995

0.99 0.9 0.985

0.98 0.85

0.975

0.8 0.97

0.965 0.75

0.96

0.955 0.7 0 2 4 6 8 10 0 2 4 6 8 10 Number of Rings Number of Rings (a)without replication (b)with replication Figure3:The experimentaland analysis result

1.04 c=1.01 c=1.01 c=1.02 c=1.02 c=1.03 1 c=1.03 c=1.04 c=1.04 1.02 c=1.05 c=1.05 0.95

1 0.9

0.98 0.85

0.8 F_latency 0.96 F_latency

0.75 0.94 0.7

0.92 0.65

0.9 0.6 0 2 4 6 8 10 0 2 4 6 8 10 Number of Rings Number of Rings (a)without replication (b)with replication Figure4:The latencyimprovementwith differentvalue of c

6Conclusions

In this paper, we presentedanapproach to make useofstructuredP2P networkfor massively multiplayeronlinegames.Chordisappliedtolocatethe game objectsmaintained by thesuppliers.Wealsoproposed an algorithmtosplit Chordtoachieve betterlookup performance. We appliedthe algorithmtoreducelookup networklatency by grouping peersthatare geographically closetoeach other. Ourevaluations in bothanalytical and experimental aspectsdemonstratethe beneﬁt of splitting.

References

[BPS06] Ashwin Bharambe, JeffreyPang, andSrinivasanSeshan. Colyseus:ADistributed Architecturefor OnlineMultiplayerGames.InProceedings of Networked Systems Design and Implementation,pages 155–168, 2006.

[BRS02] Ashwin R. Bharambe, Sanjay G. Rao, andSrinivasanSeshan. Mercury: AScalable

167 Publish-subscribeSystemfor Internet Games. In Proceedings of NETGAMES,pages 3–9, 2002. [DLS+04] FrankDabek,Jinyang Li,EmilSit, JamesRobertson, M. FransKaashoek, andRobert Morris. Designing aDHT forLow Latencyand High Throughput.InProceedings of Networked SystemsDesignand Implementation,pages 85–98, 2004. [KLXH04] Bjrn Knutsson, Honghui Lu,Wei Xu,and Bryan Hopkins.Peer-to-peer Supportfor Massively MultiplayerGames.InProceedings of IEEE INFOCOM,pages 96–107, 2004. [LSM+05] JinyangLi, Jeremy Stribling, Robert Morris, M. FransKaashoek, andThomer M. Gil. APerformance vs.CostFramework forEvaluatingDHT Design Tradeoffsunder Churn. In Proceedings of IEEE INFOCOM,pages 225–236, 2005. [LSMK05] JinyangLi, Jeremy Stribling, Robert Morris, andM.Frans Kaashoek. Bandwidth- efficient ManagementofDHT RoutingTables. In Proceedings of Networked Systems Design and Implementation,2005. [MLMB01] AlbertoMedina, Anukool Lakhina,Ibrahim Matta,and John W. Byers. BRITE: An Approach to UniversalTopology Generation. In MASCOTS,pages 346–354, 2001. [RD01] Antony I. T. Rowstron andPeter Druschel. Pastry:Scalable,Decentralized Object Location, andRoutingfor Large-ScalePeer-to-Peer Systems.InMiddleware, Lecture NotesinComputer ScienceVol.2218,pages 329–350, 2001. [RFH+01] Sylvia Ratnasamy,PaulFrancis,MarkHandley,Richard M. Karp,and ScottShenker. AScalable Content-addressableNetwork.InProceedings of SIGCOMM,pages 161– 172, 2001. [RGRK04] Sean C. Rhea, DennisGeels, TimothyRoscoe, andJohn Kubiatowicz. Handling ChurninaDHT.InProceedings of USENIX TechnicalConference,pages 127–140, 2004. [RKCD01] Antony I. T. Rowstron, Anne-Marie Kermarrec, Miguel Castro,and PeterDruschel. SCRIBE:The Design of aLarge-ScaleEvent NotificationInfrastructure. In Networked Group Communication, LectureNotes in Computer ScienceVol.2233,pages 30–43, 2001. [RMO08] Silvia Rueda, PedroMorillo,and Juan M. Orduña.AComparativeStudy of Awareness Methods forPeer-to-peer DistributedVirtual Environments. Journal of Visualization and Computer Animation,19(5):537–552, 2008. [RWF+07] SimonRieche, KlausWehrle, Marc Fouquet, HeikoNiedermayer,Leo Petrak,and GeorgCarle.Peer-to-Peer-based InfrastructureSupportfor Massively Multiplayer OnlineGames.InProceedings of 4thAnnual IEEE Consumer Communications and NetworkingConference (CCNC 2007),2007. [SMLN+03]Ion Stoica, Robert Morris, DavidLiben-Nowell, DavidR.Karger, M. FransKaashoek, FrankDabek,and Hari Balakrishnan. Chord: AScalable Peer-to-peer Lookup Proto- colfor Internet Applications. IEEE/ACMTrans.Netw.,11(1):17–32, 2003. [wow04] WorldofWarcraft. http://www.worldofwarcraft.com,BlizzardEntertainment, 2004. [ZZZ+04] Xinyan Zhang, Qian Zhang, ZhenshengZhang, Gang Song, andWenwu Zhu. A ConstructionofLocality-aware OverlayNetwork:mOverlayand ItsPerformance. IEEE Journal on Selected AreasinCommunications,22(1):18–28, 2004.

168 An Adaptive Policy Routing with Thermal Field Approach Lada-On Lertsuwanakul,HerwigUnger

Faculty of Mathematics and Computer Science FernUniversität in Hagen Universitätsstraße27-PRG 58084Hagen [email protected] [email protected]

Abstract: Thispaper introducesanadaptiverouting approach based on buffer status and distance in ameshoverlay network: Thermal Field is used for considering buffer stage and distance is applied to select routing policy. The path selectionprocess considersthe ThermalField as ametaphor forthe buffer usage basedonprobability in ordertoavoid message loss by overloaded peers or delay becauseofbig queues,otherwise routing by theshortestway.Inaddition, the probabilityofThermal Field consideration relies on leftover distance to thetarget insteadofusing aglobalconstant. Theexperimentresults substantiatedour approach using theadaptiveprobability of routing mechanismworks effectively.

1Introduction

Theboom of theinternet applicationstoday requires better support forQuality of Service (QoS). Unfortunately, thereare many factors from both, humanand technical side, causing lowinternetperformance. Theproblemsstill exist, such as packet-drop whensenttopeers with overloadedbuffers, packet-delay when residing in large queues or using indirect routes, andpacketsexpired. Thegoals of QoSrouting are usually not onlytoselect thebestpathfor sending informationfromsource to destinationeffectively but also to provideefficientnetwork utilization. Manyalgorithms [Ab07], [XG07] have been developedinthisresearch area.Someofthemare computingoptimal routes consideringtwo-ormore constrains, especially bandwidthand hop-count,but onlyafew that concernedbufferspace.

In addition to theefficiency of the algorithm,the routing performance also relies on the network-architecture. The modern internet structure emerges, e.g. in Peer-to-Peer(P2P) networksorganized in virtual community overlay-networkworking on thebasicprotocol level [Lu04].The structured overlay networksuse Distributed HashTables(DHT) to identifyarelationshipamong nodes andfilesfor routing control.Suchastructurefinds data potentially, but does not support complex lookup requests. Also it affects when any nodes leave without notification. Whereas, theunstructured types organize peers in a randomgraph or hierarchical,and use flooding or random search on thegraph to find the desiredcontent. Eachpeer queries its owncontentlocally so it supportscomplex queries.

169 Agridarchitecture is interestingbecause of Kleinberg’s work [Kl00]. He has introduced adecentralized algorithm in gridswithadded long-range links andproofeditwas able to forward messagesfromany sourcenode to targetwithin finite deliverytime. Moreover, Berg et.al. [BSU09] have shown cartesiancoordinatesystemispossibleto generateingridontop of thelarge-scaledecentralized network.

Those are ourmotivations to introduce an adaptivealgorithmfor a2-dimenstional coordinatespace overlay on unstructured P2Pnetworksusing thermal fieldtodeal with buffer stages. High temperature meanshigh buffer utilization. Theoptimal path emerges fromthe lowisotherms.Inthe first stageofresearch [LU09],the constantprobabilities set up globally were introduced.The results of simulationdemonstratedThermal algorithm that can work efficiently forconsidering buffer status.One global value, however,cannotfit to allcommunicationnetwork conditions,especiallywhenanalyzing the remaining distance. Though,the adaptive probability ideas are introducedinthis paper. Thepacketwill be forwarded by fastestpolicy or adaptivepolicy based on a probability formulawhich is thefunctionofremainingdistance from currentlocation to target node. Thecloser packets move to thetarget,the more packets use the direct path.

Thispaperisorganized as follows:Section 2discusses classical routing algorithms on meshes and thermal field algorithmsused forsearching in P2Pnetworks. Section 3 introduces an adaptiveprobability of our routing strategyusing thethermal field approach.Section4describesthe simulationenvironment P2PNetSim, respectivethe results anddiscussions. Finally, Section5concludesthe paper andgives an outlookfor futureresearch.

2Related Work

2.1ClassicalRouting on Mesh Networks

Mesh topologies have beenusedinmanyareas of communicationnetworks.The mesh network is reliableand offers redundancy of theconnection.

Themesh scheme is applicable to packet/circuit switchinginbothwireless networks [LW04],[RR91]and wired networks [CL92], [JVM95], vehicleproblemsand software interaction. The routing algorithmgenerally is theprocess to define pathsfor sending datafromanode to another through thenetwork traffic. Thegoals of routingalgorithm areproviding fastest or shortest path, preventing deadlocks, ensuring lowlatency, balancing network utilization,and fault tolerance. There are some typical routing algorithms in mesh-connected topologies [Me04].

1. Adeterministic method is called“XY routing algorithm”.Packetsroute alongX direction andchangetoYdirectionwhenreaching the Yvalue of thetarget.

2. The partial adaptive algorithms, “West-First”, “North-Last”,and “Negative- First”.Packets routewith deterministic algorithms in specific conditions;otherwise, packetsroute by using afunctionthatreacts immediately on network traffic.

170 Routing by these classical methods, there are multiple pathshavingsamehop count. The source node wantstosendinformation to the target node.Thenthe best pathdependson which algorithm is selected.The chance to find lowQoS relies on thepathselection function.Ifthe selected routehas many overloaded peers,thendelay time increases or the packetlossoccurs.

2.2ThermalAlgorithms

The Thermal Fieldapproach has beenintroduced by Ungerand Wulff [UW04]. It is used forsearching nodesinP2P networks that keep thedesired data. Thespecial information can be averyfrequently accessed data or recently update information. The temperature implies theintensity of theactivities or changes of specific informationinthe node in the web community.Further, whenahigh temperature point occurs in thecommunity, its heat spreads around. Thespreading temperaturedecreases by distance between heat source andmeasurement point, also by distribution time; the sameeffects canbe observed in the part of humanbody. Finally,apoint becomescolderifthere is no heat fedin.

Thethermal approachcan be appliedtoaP2Penvironment whenthe assumptionismade that members of the community cooperate with each others, andall peers contribute for community results.Wheneverthere is amessage sentamong members of the community, it meansthe heat is transported from source to neighbor.However,there is a difference from nature that the virtual communityisabletomemorize temperaturesfrom latest access of each neighbor.Consequently,when amessage requestsfor aspecial information, it can be transferred to the“hottest” neighbor that is kept in memory.

Section2presented some existingrouting algorithms on mesh- or grid-likenetworks. The approachesworkeffectively; however,their performance shouldbebetterifmore constraintsare considered.Further,originalthermal fieldapproach is describedfor searching specific information in P2P networks. The next section explains how the thermal fieldalgorithmworks to find routesand how thefunctionsmakethe policy selectionprocess flexible.

3Algorithms

Our approach considers buffer stages to find optimal paths. Thethermalfield is used for communicatingbufferinformationoverthe network.Sothat, every node has to keepits neighbors’ temperatures andID. Thelower temperatures represent more available resources to handlenew data. However, themainroutinggoalusually is to find the fastestway.Thenthe balance of direct way and adaptiveway must be definedproperly. Hence,the suitable probability leadstoglobalresourceutilizationand the optimal path. Basedonagrid structure,our approach uses the euclideandistance for measuring the lengthbetween nodes. In theroutedecision process, thedistance of original to target node,the length of currenttotarget peer, and the distances of neighbors to target location are measured.

171 By these results,the shortest path can be measured, andthe relative distance among interested locations can be calculated.

3.1Measuring theTemperature

In ouralgorithm,the temperature $ represents the bufferusage of apeer that is the level of messages waitingtoforward.Atacurrentnode c,the temperature $c is calculatedat every simulation time. The value of $c is between 0and 1: 0denotes an empty buffer and 1afull buffer. Messages in Buffer $ =,0≤$≤1 c Buffer size c The latest buffer status is important to make acorrect decision; hence, it is designedto attach the temperature value to alldatapackets sent through thecommunity and in the corresponding acknowledgement packets. Thepackets andthe acknowledgements work as amedian of the temperature. They pass temperatures from one to another nodeuntil they reachtheirtarget or expire.

Every current node c has aset of neighbors N(c) where messages can be forwardedto and i is anumber of neighbor,then Ni " N(c),1≤i≤4.There arethree possibilities to updateaneighbors’temperature, $ (Ni)onnode c.Let βi be thenumber of packets and μi be thenumberofacknowledgmentswhich sent from neighbor Ni to currentnode.

1. If node c receives apacketoranacknowledgment from neighbor Ni ,the old temperature is replacedwith thenew temperature.

$ (Ni)= $i,ifβi>0 and μi>0

2. If thereisnomessage sent from neighbor Np ,the newtemperature caused by the spreadofsourcenode then decreases exponentially, whereby t is therouting time.

-&t $ (Ni)=$(Ni) % e ,ifβi=0 and μi=0 3. The newtemperature is zero whennomessage arrivesand no heat remains.

$ (Ni)=0, if βi=0, μi=0, and $ (Ni)=0 Our algorithmselects therouting policy by theprobability of using thermal field.Atthe startingpoint of ongoing research, theprobabilities used forselectingthe fastest path or adaptivepaths were defined as global parameters, as presented in [LU09].The seven predefinedprobabilities are tested in theP2PNetSim simulator. All constantprobabilities showed theeffectiveresults of thealgorithm, however,neither of them fully fitsall communicationnetwork conditions.For example, when thenodewantstoforward the packet which is close to its target, thelow buffer path might be maintainedfor along routing timeinstead of forwarded directly theremainingstepswith alittle more delay. Hence,the high bufferroute with thefewer intermediatenodes has higher potentialfor routing apacket than lowbuffer pathswith longer routingtimes. Usually there is no need to useroute policies equally, so theshortest routing policy is predominantover thermalapproach in these conditions. Therefore,anadaptiveprobability forflexible

172 routing is necessary for considering distances. Next topic, we introduce five adaptive possibilities which are initiated by tuning theresults fromone experimenttoanother.

3.2Adaptiveprobability for route selection

Thepath selectionstep is related to aprobability forusing temperature data, Pθ .Each node randomly selects alow-buffer route according to Pθ,otherwise selects adirect route. If Pθ is high, that meansmorechances to select alow buffer route, then it leads to long arouting time.Onthe otherhand, asmall Pθ,raisesthe high chancetoselect a direct route, andcomes up with message loss due to overloaded nodes alongthe shortest way.Though,the optimal routeand load balancednetwork arearesultofabest probability.

Theadaptiveprobability formulas (AP$)are both linear and exponentialfunctions of relativeremainingdistance ! during therouting time t.Whenasource node s sends a packet to destination node φ,the distance betweenoriginal peer andatarget peer is ds2φ. At acurrent node c that is going to decide for apathtoforwardmessage to,the distance between currentnodeand target node is dc2φ.

Distance_from_current_to_target (t) dc2φ !(t)= = Distance_from_source_to_target (t) ds2φ 1 1 The Adaptive Probability1 (APθ ): The APθ formula is alinear function. The probability of using thermal field is on interval [0,1]bythe valueofremainingdistance, !.But the probabilityisalways equal to 1whenthe currentnodeisfarer fromtarget than theoriginal node. Thegraph of probability is presented in Fig. 1(a). !, d ≤ d 1 c2φ s2φ ProbabilityofAPθ = 1, dc2φ >ds2φ Otheradaptiveformulasare exponentialdistribution functions whichthe rateparameter (&)isequalto1in this paper.

2 2 TheAdaptiveProbability2 (APθ ): The APθ is converted fromalinear function to an exponential function. Therange of probability is [0,1], shown in Fig. 1(a), similar to 1 APθ but increases exponentially when thecurrent node is on thepathbetween source and target peer. -&(1+)1 2 !(t) Probability of APθ (!;&)= e

3 3 TheAdaptiveProbability 3 (APθ ): The APθ is theexponential probability density function (pdf) of inverse remainingdistance.The probabilityofusing thermal is on interval [0, e (-1)]when decision node is located between sourceand target node.Onthe otherhand,whenindirect routeisselected,the probability of usingthermal field is higher according to thefarerdistance. ()-& 3 !(t) Probability of APθ (!;&)= e

173 4 4 The Adaptive Probability4 (APθ ): The APθ formulaisanexponential cumulative distributionfunction(cdf). Theprobability of using Thermal field is on interval [0,1). This adaptiveideameanswhendecision node is closerthe target, theprobability of using thermal fieldishigher.

4 (-& % ! ) Probability of APθ (!;&)=1– e

5 5 The AdaptiveProbability5 (APθ ): The APθ ideaisacombination of theformulas for direct and indirect paths.Whendecision node is closertothe target than theoriginal node;the probability of thermal field decreases. Also,ifdecisionnodeisfarer away fromthe target than thesource, theprobability of using temperature decreases. -&(1+)1 e !(t) , dc2φ ≤ ds2φ 5 ProbabilityofAPθ (!;&)= ()-& (t) 1-e! , dc2φ >ds2φ

Probability Probability 1.0 1.0 0.9 0.9 0.8 0.8 0.7 0.7 0.6 0.6

0.5 1 0.5 AP$ 0.4 2 0.4 AP$ 0.3 0.3 3 AP$ 0.2 0.2 4 AP$ 5 0.1 0.1 AP$ 0.0 0.0 0.00.2 0.40.6 0.81.0 1.21.4 1.61.8 2.02.2 2.42.6 2.83.0 0.00.2 0.40.6 0.81.0 1.21.4 1.61.8 2.02.2 2.42.6 2.83.0 Relative RemaingDistance (!) Relative RemaingDistance(!) (a)(b) Fig. 1: Thegraph of adaptive probability functionswhen & =1

This section presented thedetails of theadaptiveprobabilityideafor routing with the thermal field approach to considerbuffer stages. Next in Section 4, we present some experimental results of adaptiveprobabilityfunctionscompare with predefined probability parameters that we introduced in [LU09].

4Experimental consideration

4.1Environment Setting

The simulation tool –The experiment was simulatedusing P2PnetSim anetwork simulationenvironment[Co06]. Thetoolispowerful andflexible in simulating, modelingand analyzingany kind of networks, notonlycomputer networksbut also social networks,RFID-processingand spreading of ineffectual diseases.Itisable to manage more than one million nodes. Peerscan be configured collectively and individually usingXML files for simulationsetup. Peer-Behaviorisimplemented in the Java programming language.

174 The network –Inour experiments, thenetwork is organized into agridstructurewith 10,000 nodesintwo dimensions (100x100). The coordinatesofanode within thegrid form its node ID.The gridisoverlaidonavirtual IPv4 network.Peers areconnected in four directions to eachother:left, right, up,and down.The buffersizes andoutgoing bandwidths are limited for most of thepeers. Both buffer sizes and bandwidth values are assignedrandomly followthe Paretodistribution. Thereare twotypes of packets,data packets andacknowledgements. Theacknowledgmentisprioritized.Otherwise, the system handles the packets First-In-First-Out.

The traffic pattern –Togeneratetraffic, thesimulationdefines different throughputs fornodes in terms of buffer sizes andoutgoing bandwidths.Inthe trial, the50source nodes arerandomly selected sending messages to four target nodes. They generatea messageevery 3rdsimulationstep until simulation-timehas reached300. The performance metric –Inorder to evaluatealgorithmperformances, thefollowing metrics are measured:

• number of messages loss • number of messages arrive their targets • routing timethatcounts from launching theoriginalnodetoreaching thetarget node. That time includes movingsteps andwaitingtimes in thetraffic nodes. • delay timethat summarizes from waiting timesbecause of high queuessince launched from original node until reaching the garget. • numberofnodes that have thebufferusage more than 70%. Ourassumption, this levelisthe startingpoint that cause overloaded buffer situation.

4.2. Results and Discussion

Theexperiments reported in this section compare seven globalparameterprobabilities: Fix-P0.1, Fix-P0.3,Fix-P0.4, Fix-P0.5,Fix-P0.6, Fix-P0.7,and Fix-P0.9 from simulation results in [LU09],and five adaptiveprobability functions which have been described previously. In experiments, amessage is generated andforwardedevery three simulation stepsby50sourcenodes that sent to specific four target locations.The source peers stoppedsending at simulationtime300.

Themessageisforwarded throughthe network until oneofthese cases happens: the message reaches its target, the time-to-life of the message reaches zero, or the message is deleted by an overloadedbuffer. Theexponentialrateparameter (&)isequal to 1.

Fig.2showsthe comparison of theratio with threekey performance metrics: messages arrivetheir target, messages lose due to overloadbuffers, and messages expire before 5 they have found theirtargets.The APθ shows thebest resultamong theadaptiveones, 88%reached target and 12% lost. However, afixedprobability of 0.6 presentsbetter results which is 100% reachtheir targetsand no message lose andexpire. In contrast, 1 2 APθ andAPθ are inefficient to reach thetarget. About 40% of all messages reachthe goal,but others lose andexpire. By theway,Fix-P0.9 is theworst,only6%ofthe messages areabletoreachtheir target, and93% expire.

175 Our assertioniswith too high probability (e.g.Fix-P0.9)and toolow probability (e.g. 1 2 Fix-P.0.1, APθ , APθ )are ineffective, theoptimal routecan be found by balancing both 4 policies (e.g. Fix-P0.5, APθ ).

MessageArrived MessageLost MessageExpired 99% 100% 93% 93% 88% 88% 90% 84% 82% 84% 80% 76% 75%

70%

60% 53% 50% 39% 37% 40% 33% 30% 30% 24% 25% 18% 20% 15% 13% 12% 12% 8% 10% 7% 6% 3% 0% 0% 0% 0% 0% 0% 1% 0% 0% 1% 0% AP1AP2 AP3 AP4 AP5 FixP0.1Fix P0.3 FixP0.4Fix P0.5 FixP0.6Fix P0.7 FixP0.9 Probability Functions Fig. 2: Thecomparison of ratio message arrivedtarget, message lost,and message expired

Fig. 3and Fig. 4analyzethe number of messages reachedtheir targets during simulation time. Both results of Fix-P0.9 show very low routing time and delay time because the number of messages could reach their targets in 6%.Fig.3presentsthe average of 5 routing timecompared to the average of shortest paths. Althoughthe APθ performs effectivebyreaching theirtargetwith high success; it surprisinglyhas very high routing time,200 timescomparedtothe shortest routewhich is causedbylongroutesfrom 3 bufferusage consideration. The APθ shows thebest routing time among adaptive probabilities.

Simulation Time 1,200 Avg. RoutingTime 1,091 Avg. Shortest Path 1,000

800

600 439 400 286 267 293 176 190 180 200 137 83 105 58 58 55 58 54 73 55 57 58 58 59 53 18 0 AP1AP2 AP3AP4 AP5Fix P0.1 FixP0.3Fix P0.4 FixP0.5Fix P0.6 FixP0.7Fix P0.9 ProbabilityFunction Fig. 3: Theaverage of routing time compare to average of shortest path

Fig.4presents theaverageofdelay time which summarized fromwaitingtimedue to high buffer queue. Almostadaptiveprobabilities have low delay time,and lower than all 3 5 constant probabilities, except APθ which similar to Fix-P0.5. Especially APθ hasthe lowest delay time, furtherithas ahigh number of messages reaching their target.

The next diagram,Fig.5,presentsthe amount of high buffer usagenodes in the community when thefractionofbuffer usageisover70%.The adaptiveprobability 5 demonstrated remarkably results. The APθ shows excellent; there is no node has higher level of buffer queue than 70%, andthe buffers of three nodesare filled to 70% only. In 1 2 contrast, APθ and APθ have many high buffer usage nodes;furthermore, they arehigher than all constantprobabilities.

176 AverageDelay Time (simulationtime) 15 14.02

12 11.06 9.58 9.71 9.11 9 7.38 7.59 6.66 5.70 5.88 6 5.21

3 0.69 0 AP1 AP2AP3 AP4AP5 FixP0.1Fix P0.3 FixP0.4Fix P0.5 FixP0.6Fix P0.7 FixP0.9 Probability Function Fig. 4: Theaverage of delaytime

Number of Peers 600 FULL(100%) 90% 80% 70% 553

500

400 381 335

300 255

208 202 200 184 132 116 103 91 100 80 67 61 63 44 44 45 51 43 46 44 34 26 30 31 29 33 32 26 36 27 18 19 18 18 19 19 16 12 0 0 0 3 3 5 9 9 0 AP1 AP2AP3 AP4 AP5Fix P0.1 FixP0.3Fix P0.4 FixP0.5Fix P0.6 FixP0.7Fix P0.9 ProbabilityFunction Fig. 5: Thesummaryofnumber of high buffer usage nodes (70%up)

Fromthese simulationresults, it is clear that there is no probability that shows outstanding in every simulationscenarios andall performance metrics. Aconstant probabilityof0.6 is the best according to the numberofmessagereaching their target but it takes high routing time and high delay time.The number of messages reaching their target withconstantprobability of 0.5isworse but routing time anddelaytimeis better compared to thesimulationwith probability of 0.6. Theresultsofadaptive probabilities perform better than global constantprobabilities in balanced resource 5 utilization. The APθ shows good results;nohigh bufferusage nodeand lowwaiting 4 time,howeverittakes very long routingtime. The APθ hassimilar behaviortothe fixed probabilities of 0.4and 0.5but theadaptiveone haslower delay time.

In Fig. 6, thebuffer utilizationstatusofthe community(100x100) is captured for presenting thealgorithm’sperformance. Thesequence of picture is read from left to right. The application froze thebufferusage status at thesimulationtime-steps 150, 300, 450, 600,750 and900.The 50 source nodes arerandomlydistributed in thenetwork. The four target locations areclose to each other in theright-down corner. The density of thecolor presents bufferlevel of node. Thehigher bufferusage levelsare shownas 1 darker colors. Theprobability of APθ decreaseswhen thecurrentnodeisclosed to the target but its probability is always 1whenthe currentnodeisinanindirect route. In Fig. 6(a), messages are distributed over thenetwork when simulationstarted.After that, messages between source andtarget could reachtheir target, but whenmessages are forwarded more indirect way then many grey spotsonthe top-left of thepicture are shown.

177 Time: 150 300 450 600 750 900

1 (a) Adaptive probability function APθ

3 (b) Adaptiveprobability function APθ

5 (c) Adaptive probability function APθ Fig. 6: Thebuffer status diagramsofacommunity 10,000 peersin2-dimensions (100x100)

3 In contrast, the APθ probabilityrange is lower. Then almostmessages are forwarded with shortest policy. They couldreach their target quickly. Hence,diagramsattime 150 and 300onFig.6(b)havedarkgrey lines thatrepresent high congestionnodes,and from 5 time 450, thereisnomessage left.Lastly, the APθ which has twofunctions in different criteria. Thediagram,Fig.6(c) hasnodarkgrey nodes but there are many light grey 5 nodes spreading over thenetwork.Itconfirmsthe simulationresult of APθ that hasno high bufferusage.

5Conclusion and Future Work

In this paperweintroduced theadaptiveprobabilities usingthermalfield approach considering relativeremaining distance.The experimentsrun with apowerfulsimulation tool,P2PNetSim. Thetest results proofthatthe thermal field algorithmenabled to find an appropriatepath, andreact to high buffer usage situations. Butwith differences performance among probabilities of using thermal field algorithm can be understoodon thebasis of the differentdegreeofadaptively which thedifferentflexibility respondto distance changing in time.

In future work, more constraints,suchasbandwidth will be considered for improving quality of service routing.Inaddition,multi-criteria have concurrently to be considered to providemoreefficientglobal routing optimization. Finally, theenhancementof routing algorithmswill be studied by learning process.

178 References

[Ab07]Abraham, A.; Yue, B.; Xian,C.; Liu,H.; Pant, M.: Multi-objectivePeer-to-Peer Neighbor-Selection Strategy Using Genetic Algorithm: LNCS 4873, 2007; S. 443-451. [BSU09]Berg, D.;Sukjit,P.; Unger,H.: Grid GenerationinDecentralized System,2009. [CL92] Cristopher, J.G.; Lionel, M.N.: Adaptive RoutinginMesh-Connected Networks: Proc. 12th International Conference on Distributed Computing Systems, 1992; S.12-19. [Co06] Coltzau, H.: Specification and Implementation of Parallel P2P Network Simulation Environment: DiplomaThesis, University of Rostock, 2006. [JVM95] Jatin, H.U.; Varavithya,V.; Mohapatra, P.: Efficientand BalancedAdaptive Routing in Two-DimensionalMeshes: In InternationalSymposium on High Performance Computer Architecture,1995; S. 112-121. [Kl00] Kleinberg,J.: The small-world phenomenon: An algorithmic perspective: Proc. 32nd ACMSymposium on Theory of Computing, 2000. [Lu04] Lua, E.K.; Crowcroft, J.; Pias, M.; Scharma, R.;Lim,S.: ASurveyand Comparison of Peer-to-Peer OverlayNetwork Schemes: IEEE Communications Survey and Tutorial, March 2004. [LU09] Lertsuwanakul, L.;Unger,H.: AThermal Field Approach in AMesh Overlay Network: 2009. [LW04] Lee,A.; Ward, P.A.S.:AStudyofRouting Algorithms in WirelessMeshNetworks: Australian TelecommunicationNetworks and Applications Conference, December 2004. [Me04] Mello, A.V.; Ost, L.C.;Moraes F.G.; Calazans N.L.:EvaluationofRouting Algorithms on Mesh Based NoCs: Technical Report Series No.040,May 2004. [RR91] Rajasekaran,S.; Raghavachari, M.:Optimal Randomized Algorithms for Multipacket and Wormhole Routingonthe Mesh: Technical Report,University of Pensylvania, 1991. [UW04] Unger, H.; Wulff, M.: Search in Communities: An Approach Derivedfromthe Physic Analogue of Thermal Fields: Proc. the ISSADS 2004, LNCS 3061,Guadalajara, Mexico, 2004. [XG07]Xu, M.;Guan, J.:Routing Based Load Balancing for Unstructured P2P Networks: FGCN-Future Generation Communication and Networking (FGCN 2007) -Volume 1, 2007;S.332-337.

179

Session6

Semantic Web Technologies

From Community towards Enterprise –ataxonomy-based search forexperts

Gerald Eichler∗,Andreas Lommatzsch†,Thomas Strecker†, Danuta Ploch†,ConnyStrecker†,Robert Wetzker†

∗Innovation Development Deutsche Telekom AG,Laboratories Deutsche-Telekom-Allee 7 D-64295 Darmstadt [email protected]

†DAI-Labor,Technische Universita¨tBerlin Ernst-Reuter-Platz 7 D-10587 Berlin {andreas.lommatzsch|thomas.strecker|danuta.ploch| conny.strecker|robert.wetzker}@dai-labor.de

Abstract: In this paper we introduce aversion of the Spree expert finding framework [BAA+07] tailored for enterprises. Whereas expert finding services have been very successful on the Web, enterprise levelsolutions are still scarce. This comes as asurprise, as the process of finding the right person (to ask) among colleagues requires aconsiderable percentage of most employees’ time yielding ahigh potential for optimization. The core of Spree is an expert finding algorithm that automatically maps questions to the most qualified experts usingadomain-specific topic taxonomy as intermediate. Apart from the framework itself, we describe the challenges and design decisions that have to be taken into consideration when implementing expert finding solutions in enterprises. These include the selection of an appropriate domain taxonomy,the motivation of employees to share their knowledge and privacy related concerns.

1Motivation

The classical Internet provides information in the form of documents. Users searching for information can either directly access these documents or use search engines to identify the most relevant ones for agiven query.However,documents contain only afraction of the entire knowledge. In some cases, instead of finding adocument, the user might want to find the right person to ask. Whereas the need for expert finding solutions has originated

183 184 Questionprocess

Ask question Reclassify Select expert(s) StartChat

Questioner Expert profiles Expert(s)

Rate expertise Read blog Annotateblog Create blog

Blog knowledge base

Figure 2: The question and blog process flows. tures improve classification quality compared to flat approaches (e.g. [KS97], [DC00]). The current implementation of the matching logic does not support typed categories, such as categories for locations in contrast to skill categories, or the assignment of additional category attributes. However, in future versions of Spree, we plan to integrate these concepts well known from ontologies in order to allowfor amore complexknowledge design. In this paper,wewill therefore not distinguish between the words taxonomy and ontology. Experts are informed wheneveranew question within their knowledge domain appears and can answer aquestion in real-time using the provided chat or email functionality.Ad- ditionally,users of Spree are encouraged to create blog entries about topics theyconsider interesting for other community members. These entries together with all questions and answers can be searched and rated by the community increasing the effectiveness of the framework overtime. Figure 2depicts the knowledge generation process. This paper is structured as follows: Section 2describes the Spree framework and its core components. We then outline the challenges related to the design of an enterprise knowledge taxonomy and present the ontology editor tool that helps during the taxonomy construction process. Section 4presents the expert finding algorithm, the core functionality of the Spree framework. We then discuss the challenges we met during the development of Spree and conclude with an outlook on possible future improvements.

2The Spree framework

Forthe development of Spree, amodular approach waschosen. This modularity allows the reuse of components, such that e.g. the expert finding functionality can be integrated into other possibly existing solutions. The Spree framework consists of six functional components: User Management, Ontology Management, Request Management, Matching, Communication and Community.Wedescribe the Spree components and their interaction on the example of the ask process which includes the classification of questions as well as the expert finding and communication (Figure 3).

185 3b.Find expertise Ontology Management 1b. Find ontology Matching profilesfor User Management classified request Creation, Import, Modification nodes for request Preprocessing, Request Registration,Deregistration, Classification,Expert Personal Profile, Expertise 2b. Get ontology Identification nodes

1a. Classify request, 2a.Reclassify request 3a. Find experts for (classified) request 6b. /8b. 4a. Getpreferences Adapt expertise forexperts profile to rating RequestManagement Creation, Conversion, Forwarding 4b. Contact selected experts 5a. Initiate conversation

0. Insertrequest Communication Community Chat, E-Mail Blogs, UserStatistics, Rating 5c. Chat 6a. Rate communication

5d.End conversation 7. Create blogentry from request Consumer

Experts Users

8a. Rate 5b. Join communication blog entry 5c. Chat

Figure3:Active components within the ask process ﬂow.

The central component within the ask process is the Request Management that manages the entire life cycle of aquestion including the creation of the corresponding request objects and its forwarding to the most appropriate experts. The component also handles user actions referring to requests, such as view, delete or close actions. In order to find the best matching experts, the request has to be classified (step 1). Both the classification and the subsequent expert matching are handled by the Matching component. This component performs alanguage-dependent analysis of the request’stextual representation and classifies the request to ontology nodes. The resulting classification is then presented to the user who can modify it if required. Forthe expert matching process, the matching component also interacts with the Ontology Management that provides the current ontology and ontology related functionalities, as well as with the User Manage- ment where user expertise profiles, user preferences and previously receivedratings are stored. The resulting expert list is then presented to the questioner who can add or remove experts. This modification step may be required in cases where the questioner already knows about potential expert candidates or wants to exclude certain experts from the process. The Request Management then forwards the request to the selected experts (step 4). If an expert is currently on-line, he will be informed in real-time about the incoming question. However, experts do not always have to be logged in butmay also be informed about newrequests via email if specified in their user preferences. The Request Management then triggers the initialization of aconversation between questioners and experts viathe Communication component (step 5). Experts can reply to arequest by sending an email

186 that will be forwarded to the questioner or,inthe majority of cases, by joining achat started for each request. After aconversation has been ﬁnished, the consumer can rate it (step 6a) and all participants are asked to summarize the conversation in ablog entry (step 7) that may be found by potential future questioners. All rating and blog functionality is handled by the Commu- nity component. In addition to conversations, also blogs (step 8a) can be rated. An expert can thus improve his score by providing quality answersorbywriting blog entries and augmenting the Spree knowledge base. The Community component is also responsible for updating the expertise proﬁles based on receivedratings (steps 6b, 8b). Furthermore, the community component provides all functionality for presenting system statistics, e.g. high score lists or an overviewabout the most popular topics. The statistical data is designed to motivate community members to share their knowledge.

3Taxonomy design

Designing adescriptive taxonomy is crucial for the success of the expert finding solution. Design decisions do not only include the specification of relevant categories and their arrangement into ahierarchical structure, butalso require thoughts on the type of classification alater classifier should return and on the optimal taxonomy size. Forthe category selection, it is crucial to define what expertise domains are essential. Here, expertise may be seen from askill perspective where employees are considered experts based on their knowledge about technologies, products, business partners or markets. However, in some cases the taxonomy may simple reflect an enterprise’sinternal structure considering users as experts if theyare responsible for acertain task or domain. The Spree approach allows unifying the skill perspective and amore structural viewinto asingle taxonomy.This is made possible by the fact that Spree uses amulti-class classifier that categorizes texts to multiple branches of the taxonomy. The selection of an optimal taxonomy size is also crucial. Even though the granularity of alarge-scale taxonomy may coverall topics relevant for agiven domain, the size of such ataxonomy remains aconsiderable obstacle both for classification and usability.Onthe other hand, small taxonomies are easiertodesign, result in better usability,but may not provide enough granularity to fully describe adomain. The number of categories should generally increase with the number of experts in the system. To solvethis problem, Spree includes an ontology editor tool for the manual tailoring of taxonomies during run-time. An automated approach to the problemoftaxonomy tailoring is described in [WUH+08]. Using the ontology editor,each topic node can be defined by aname and aset of meta-data, such as adescription and aset of characteristic keywords. To each category,the editor can then assign documents that he considersdescriptive.These documents are converted to plain text and analyzed based on natural language processing tools (see section 4.1). In a next step, the ontology editor creates an n-gram frequencystatistic that will later be used to train the Spree text classifier.The editor also supports the modification of the taxonomy structure itself by adding neworremoving existing categories. Furthermore, it is possible

187 Figure 4: Screenshot of the Spree Ontology Editor shows the topic hierarchy(on the left side), the meta data and the visualization of the term statistic for the selected node. to move categories or entire branches for more complexstructural changes. Taxonomies designed with the ontology editor can be exported to anew Spree instance or to aﬁle archive.The tool also allows making changes to the taxonomy of arunning system which may be arequirement for dynamic domains. The ontology editor interface is shown in ﬁgure 4. An enhancement planned for the near future is the automated suggestion of subcategories for topics, e.g. by using acluster algorithm such as the one described in [Bis06] to group the documents assigned to atopic.

4Finding the right experts

One of the core functionalities of the Spree framework is its matching algorithm. This algorithm classifies agiven question to the topics of apredefined taxonomy and identifies the most qualified experts based on their expert profiles within the same ontology.Once the taxonomy has been designed and relevant documents and keywords have been assigned using the ontology editor,the Spree text classifier is trained. This classifier allows us to map agiven question or document to the most relevant categories. Based on these

188 Expert Request Documents, Websites, Title, Question Keywords

Preprocessing Ontology Preprocessing

Term1 Frequency1 Term1 Frequency1 Term2 Frequency2 Term2 Frequency2 … …

Termn Frequencyn Termn Frequencyn

Classification Classification

Weighted Subtree Weighted Subtree

Similarity

Figure 5: The Spree expert matching process that calculates the similarity of questions and experts using an ontology as intermediate. categories, we can then identify the most qualiﬁed experts. Figure 5gives an overviewofthe Spree expert matching process which we will describe in detail in the consequent sections.

4.1 Text processing

The text processing task for the transformation of documents and questions is subdivided into multiple steps. First, atextparser splits the text input into tokens. The sorted token list is then used to extract the text n-grams where the parameter n can be set in the system parameters. Next, all n-grams that consist of stop words are removedfrom the n-gram set as theytransport little information. The remaining n-grams are stemmed using the snowball stemming algorithm4 and aggregated to n-gram histograms that are considered theinter- nal representation of an input text. The text processing algorithm is applied identically to documents assigned to categories in the ontology editor as well as to user questions. The current implementation of the algorithm supports German and English texts.

4http://snowball.tartarus.org/

189 4.2 Text classiﬁcation

The ontology editor assigns keywords and descriptive documents to each node of the ontology.These keywords and documents are transformed to n-gram distributions during the text processing step such that each topic node can be represented by acharacteristic n-gram distribution (vector). As the provision of sufficient training documents and keywords may require avery high manual effort, it is also possible to learn characteristic n-gram patterns by using external sources such as Intranet or Internet search engines [WAB+07]. How- ever,this automatic retrievalprocess may not produce satisfying results in domain specific scenarios with their ownvocabulary. The Spree framework is designed to support avariety of classifier types. The selection of aclassifier type may depend on the number of available training material or the required classification speed. By default Spree will use aNa¨ıveBayes text classifier.For each node, this classifier estimates the likelihood that the corresponding n-gram distribution generated the n-gram sequence observed in the input text. The m most likely nodes are then considered valid classifications where m is asystem parameter.Classifications are always complete in the sense that if acategory is assigned to agiven text also all parent categories are considered valid classifications. Classifications, therefore, always appear as subtree of the taxonomy.

4.3 Expert matching

The Spree expert matching algorithm identiﬁes experts to agiven question based on the ontology tree T whose Nodes N = n1,...,nN correspond to the different knowledge areas. Apart from the structural assumption, the Spree system remains independent of the nature and content of the ontology considered in anyimplementation. The fundamental idea of the matching algorithm is to represent experts and user questions as serialized vectors of nodes v(T ) ∈ S(T ) where S(T ) ⊂ RN is the ontology space. The values of v(T) are set to 0or1.Once all registered experts e1,..., eE andanincoming question q have been mapped to subtrees, it is possible to compute the similarity between an expert and the question by calculating the weighted dot product

score(q, ei)=v(q)Wv(ei) where W represents aweight matrix that allows us to incorporate further contextual information about the topics of the underlying ontology.Aquestion is then forwarded to the experts with the highest score. Adetailed description of the Spree matching algorithm is givenin[BAA+07].

190 5Discussion of practical experiences

Wheneveranew application or tool is introduced in alarge enterprise, several aspects have to be taken into consideration. Most time consuming, at least in Germany, is getting the agreement of the workers’ council, as anyhandling of personal data requires its agreement. To convince all involved parties, clear concepts are needed from the very beginning covering the following items:

• Data protection concept to fulﬁll legalissues • System security concept to meet basic IT requirements • Operational concept to run the application • Business plan to convince the upper management • User guidelines to ﬁnally reach the people

Community tools on the Webwork because of the voluntary contribution of their users. For enterprises, people have to see the advantages of their participation, too. As known from the e-learning and knowledge domain, even though incentivesare often under discussion, butpublic scores that help to motivate users and increase trust in most public scenarios are often rejected as devil’swork. The advantages and disadvantages of expert scores have to be evaluated for each possible domain separately and then communicated to the management. Despite these incentives, general software quality requirements are essential for success. Especially usability aspects are of crucial importance. Users have to easily understand the application’sscope and functionality.This requires the application design to followthe design guidelines also found in existing applications as well as apossible integration into existing solutions. Compliance requirements generally include:

• Compliance with the standard IT workplace: hardware, software versions, security and browser settings, user execution rights, etc. • Compliance with the IT production platform: virtual servers, database environments, data protection guidelines, etc. • Compliance with the enterprise user administration: LDAP authentication, single- sign-on, password guidelines, etc. • Compliance with corporate design: (Web-)design guidelines, possible integration into existing portals, etc.

User satisfaction also depends on the quality of answers and time for receiving aresponse. The quality of the expert proﬁles and the underlying matching algorithm is therefore crucial. Also, to overcome the well-known cold start problem of community tools, it is necessary to motivate asufﬁcient participation rate especially in the early phase.

191 Spree can be seen as aﬂexible process overlay,not as areplacement. Therefore, interest conﬂicts can occur,giving the management the feeling that working time is stolen. How- ever,the savings are much bigger by avoiding double work, and the staffsatisfaction index will increase. Keep always in mind,the knowledge of people is the most valuable thing a companyhas.

6Outlook

There exist avariety of possible directions for further improvements of the Spree enterprise solution: Automatic profile creation: There exist manysources for the automatic creation of employee expertise profiles. One very promising source are the documents an employee has created or worked with. The topics of these documents could be used to estimate an initial expertise profile. This idea is especially appealing as enterprises generally possess acentral file storage system. Foranexample on automatic expert finding based on documents see [SHFA07]. Dynamic profile updates: If auser successfully answers questions in areas not listed in his expert profile, he should be asked to update his profile. The same should happen if he has authored documents related to other topics. Learning classifier: Users give feedback when theyaccept or modify aproposed classification during the ask process or the creation of ablog entry.This feedback is avaluable source for acontinuous improvement of the classification accuracy. Visualizing the enterprise knowledge graph: Tools that help to visualize the knowledge structure and diffusion processes within an enterprise provide an invaluable source for management decisions. The right visualization may also guide users during the process of expert(-ise) finding, as done e.g by the SkillMap tool5,and emphasize areas where an enterprise lacks domain knowledge or skills. The community version of Spree already provided avisualization tool, the Spreegraph6,for browsing the social and knowledge graph. Forfuture work, we plan to extend this existing solution toward the Enterprise 2.0 scenarios. Limit effects of malicious behavior: Whereverusers can rate other users, there exists an incentive to cheat. The effect of cheating should be reduced, e.g. by allowing users to rate other experts only once. Integration with existing Document Management Solutions: To allowquestions and discussions about certain documents, there should be atight binding between Spree and existing document managementsolutions (DMS).

5http://ioe-skillmap.hu-berlin.de/ 6http://www.askspree.de/static/ﬂash/SpreeMainComponent.html

192 References

[BAA+07] Christian Bauckhage, Tansu Alpcan, Sachin Agarwal, Florian Metze, Robert Wetzker, Milena Ilic, and Sahin Albayrak. An Intelligent Knowledge Sharing System for Web Communities. In Proceedings of the IEEE International Conference on Systems, Man and Cybernetics 2007.IEEE Computer Society Press, 2007.

[Bis06] Christopher M. Bishop. Pattern Recognition and Machine Learning (Information Sci- ence and Statistics).Springer,August 2006.

[DC00] Susan Dumais and Hao Chen. Hierarchical classiﬁcation of Webcontent. In SIGIR ’00: Proc. of the 23rdannual int. ACMSIGIR conf.onResearch and development in information retrieval,pages 256–263, NewYork, NY,USA, 2000. ACMPress.

[KS97] Daphne Koller and Mehran Sahami. Hierarchically Classifying Documents Using Very FewWords. In ICML ’97: Proc. of the 14th Int. Conf.onMachine Learning,pages 170–178, San Francisco, CA, USA, 1997. MorganKaufmann Publishers Inc.

[SHFA07] PavelSerdyukov,Djoerd Hiemstra, Maarten Fokkinga, and Peter M. G. Apers. Gener- ative modeling of persons and documents for expert search. In SIGIR ’07: Proc. 30th int. ACMSIGIR conf.onResearch and development in information retrieval,pages 827–828, NewYork, NY,USA, 2007. ACM.

[WAB+07] Robert Wetzker,Tansu Alpcan, Christian Bauckhage, Winfried Umbrath, and Sahin Albayrak. An unsupervised hierarchical approach to document categorization. In WI ’07: Proceedings of the IEEE/WIC/ACM International Conference on WebIntel- ligence,pages 482–486, Washington, DC, USA, 2007. IEEE Computer Society.

[WUH+08] Robert Wetzker,Winfried Umbrath, Leonhard Hennig, Christian Bauckhage, Tansu Alpcan, and Florian Metze. Tailoring Taxonomies for Efﬁcient Text Categorization and Expert Finding. In WI ’08: Proceedings of the IEEE/WIC/ACM International Conference on WebIntelligence,Sydney, Australia, 2008. IEEE Computer Society.

193 Recommending Related ArticlesinWikipediavia

aTopic-Based Model

Wongkot Sriurai, Phayung Meesad, Choochart Haruechaiyasak

Department of InformationTechnology Faculty of InformationTechnology King Mongkut's University of Technology NorthBangkok (KMUTNB) 1518 Pibulsongkarm Rd., Bangsue,Bangkok 10800

DepartmentofTeacher Training in Electrical Engineering FacultyofTechnical Education King Mongkut's University of Technology NorthBangkok (KMUTNB) 1518 Pibulsongkarm Rd., Bangsue,Bangkok 10800

Human LanguageTechnology Laboratory(HLT) National Electronics andComputer TechnologyCenter(NECTEC) Thailand Science Park,Pathumthani12120,Thailand

s4970290021@ kmutnb.ac.th [email protected] [email protected]

Abstract: Wikipedia is currentlythe largestencyclopedia publicly availableonthe Web. In additiontokeyword searchand subject browsing,users may quickly accessarticles by following hyperlinks embeddedwithin each article. Themain drawback of this method is that some linkstorelated articles could be missing from the current article.Also, arelated article could not be inserted as ahyperlink if there is no termdescribing it within the current article.Inthis paper, we propose an approach forrecommending relatedarticles based on the LatentDirichlet Allocation(LDA) algorithm. By applyingthe LDAonthe anchor textsfrom each article,asetofdiversetopics could be generated.Anarticle can be representedas aprobability distribution over this topic set. Two articles with similar topic distributions areconsidered conceptuallyrelated.Weperformed an experimenton the Wikipedia Selection for Schools which is acollectionof4,625selected articles from the Wikipedia. Based on some initial evaluation, our proposed method could generate aset of recommended articles which aremorerelevantthan the linked articles given on the test articles.

194 1Introduction

Wikipediaisawell-known free-content encyclopedia written collaboratively by volunteers andsponsored by thenon-profitWikipedia Foundation1.The aim of the project is to develop afree encyclopedia for manydifferent languages.Atpresent,there are over 2,400,000articlesavailableinEnglish andmanyinother languages. Thefull volumeofWikipedia contents,however,containssomearticleswhich are unsuitablefor children.InMay 2007,the SOS Children's Villages, theworld's largest orphancharity, launched theWikipedia Selectionfor Schools2. Thecollectioncontains 4,625selected articles based on theUKNationalCurriculumand similarcurricula elsewhere in the world. All articles in thecollectionhavebeencleaned up andchecked forsuitability for children.

The contentofWikipedia forSchoolscan be navigated by browsing on apictorial subject index or atitle word index of all topics. Table 1lists thefirst-level subject categories available from the collection. Organizingarticles into thesubject category set provides users aconvenient way to access the articles on thesamesubject. Each article contains many hypertext links to other articleswhich arerelatedtothe currentarticle. However,the links which were assigned by theauthors of thearticle cannot fully cover all related articles. Oneofthe reasons is due to thefact thatthere is no term describing relatedarticles within thecurrentarticle.

Table1:The subject categories under theWikipedia Selection forSchools.

CategoryArticles CategoryArticles Art 74 Business Studies88 Citizenship 224Countries 220 Design andTechnology250 Everydaylife380 Geography 650History 400 IT 64 Language and literature 196 Mathematics45Music 140 People680 Religion 146 Science 1068

1Wikipedia. http://en.wikipedia.org/wiki/WikiPedia

2Wikipedia Selectionfor Schools. http://schools-wikipedia.org

195 Somepreviousworks have identifiedthisproblem as themissing link problem and also proposedsomemethods for automatically generatinglinkstorelated articles. J. Voss [Vo05] presented an analysis of WikipediasnapshotonMarch 2005.The study showed that Wikipedialinks form ascale-free network andthe distribution of in-degree andout- degreeofWikipediapagesfollows apower law. S. Fissaha Adafre andM.deRijke [FR05]presentedanautomated approachinfindingrelated pages by exploring potential links in awikipage.They proposed amethod of discovering missing links in Wikipedia pages viaaclustering approach.Theclustering processisperformedbygrouping topically related pages usingLTRankand then performingidentificationoflink candidates by matching theanchor texts. Cosley et al.[Co07]presented SuggestBot, software that performsintelligenttask routing (matching people with tasks) in Wikipedia. SuggestBotusesbroadly applicablestrategiesoftextanalysis,collaborative filtering, andhyperlink following to recommendtasks.

In this paper, we propose amethod for recommending related articlesinWikipedia basedonthe LatentDirichlet Allocation (LDA)algorithm.Weadopt thedot product computationfor calculatingthe similaritybetweentwo topic distributionswhich represent twodifferent articles. Using theproposedapproach,wecan find therelation betweentwo articles and use this relation to recommend links for each article. The rest of paperisorganizedasfollows.InSection2,wedescribe thetopic-based mode for article recommendation. Section3presents experimentsand discussion.Finally, we conclude our work andput forwardthe directions of our future work in Section4.

2The Topic-BasedModel for Article Recommendation

There have beenmanystudies on discovering latent topics from text collections [SG06]. LatentSemanticAnalysis (LSA) uses singularvalue decomposition(SVD) to maphigh- dimensionalterm-by-document matrix to alower dimensional representationcalled latent semantic space [De90].However, SVD is actually designedfor normally- distributeddata. Such adistribution is inappropriatefor count datawhich is what aterm- by-document matrix consistsof. LSAhas been appliedtoawide varietyoflearning tasks, suchassearchand retrieval [De90] andclassification [Bi08].AlthoughLSA have achieved importantsuccess but LSA have some drawbacks such as overfittingand inappropriate generativesemantics [BNJ03].

196 Due to thedrawbacks of theLSA, theLatentDirichlet Allocation(LDA) hasbeen introduced as agenerative probabilistic modelfor aset of documents[BNJ03].The basic idea behind this approach is that documentsare represented as random mixtures over latent topics. Each topic is represented by aprobability distributionoverthe terms. Each articleisrepresented by aprobability distribution over thetopics. LDA hasalso been appliedfor identificationoftopics in anumberofdifferentareas. Forexample,LDA has been used to find scientific topics fromabstractsofpapers publishedinthe proceedings of thenational academyofsciences [GS04]. McCallumetal. [MC05] proposedanLDA- basedapproachtoextract topicsfromsocial networksand appliedittoacollectionof 250,000Enron emails.Newmanetal. (2006) appliedLDA to derive 400topicssuchas Basketball, Harry Potterand Holidaysfromacorpus of 330,000 New York Timesnews articles and represent eachnewsarticle as amixture of thesetopics [Ne06].

Haruechaiyasak andDamrongrat [HD08]appliedthe LDAalgorithm forrecommending related articlesinWikipedia Selectionfor Schools, however,withoutproviding any comparative evaluation.

Figure 1: TheLatentDirichlet Allocation (LDA) model

Generally, an LDA modelcan be represented as aprobabilistic graphical modelas shown in Figure 2[BNJ03]. There are three levels to theLDA representation.The variables α and β are thecorpus-level parameters, which are assumedtobesampled during theprocess of generatingacorpus. α is theparameter of theuniformDirichlet prior on theper-documenttopicdistributions. β is theparameter of theuniform Dirichlet prioronthe per-topicworddistribution. θ is adocument-level variable, sampled once per document. Finally, thevariables zand ware word-levelvariables and are sampled once for each word in eachdocument.The variable Nisthe numberofword tokens in adocumentand variableMis thenumber of documents.

TheLDA model[BNJ03]introduces aset of Klatent variables, called topics. Each word in thedocumentisassumed to be generated by one of thetopics. The generativeprocess foreachdocument wcan be described as follows:

197 1. Choose θ ~Dir ()α:Choose alatenttopics mixturevector θ from the Dirichlet distribution.

2. For eachword n ∈ ww

(a) Choose atopic zn ~Multinomial ()θ:Choosealatenttopic zn fromthe multinomialdistribution. (wp |z β), (b)Choose aword wn from n n, amultinomialprobability

conditionedonthe topic zn .

In this paper, we focusonthe WikipediaSelectionfor schools forevaluating our proposedrecommendationalgorithm. Our proposed approach basedonthe topic model forrecommending related articlesand discovering missing links consists of three main processes as showninFigure2.

Figure 2: Theproposed topic-based model via LDAalgorithm for article recommendation.

1. Extract anchor-textlinksfromall 4,625 Wikipedia Selectionfor School articlesand store anchortexts in thedatabase. 2. Prepare article titles and anchor texts from previousprocess as theinput to generatethe topic mode basedonthe LDAalgorithm. The output fromthis step is the topicprobability for each article. 3. Thearticle similarity is computed by using thedot productbetweentwo topic probability vectors.The scoresfromthe dot-product calculationare used to rank thetop-10articles that are related to thecurrentarticle.

198 The process forrecommending related articlescan be explained in details as follows. Theinput datafor theLDA algorithmconsistsofadocument corpus. In this paper, we present each article with the title and anchortexts. The corpus is aset of mdenotedby

= {}0,...,ddD m−1.Eachdocument is aset of ntopics denotedby i = {}0,...,ttd n−1.

Finally, each topicisaset of distribution over pwords denotedby i = { 0,...,wwt p−1}.

To recommendrelated articles, we calculate thesimilaritybetweenagivenarticle and all otherarticlesand select theoneswith thehighest similarityvalues.Given two articles i i = { j ,...,ttd j } representedasthe topic distribution vectors, i = { 0 ,...,ttd n−1} and j 0 n−1 ,the dot product canbecalculated as follows.

n−1 jiji i j . ji ∑ ji 2211 ...+++== nttttttdddd n i=0

3Experiments and Discussion

TheWikipedia Selectionfor Schoolsisavailable fromthe SOS Children's Villages Web site3.Weusedthe LDA algorithmprovidedbythe linguistic analysis tool called LingPipe4 to runour experiments. LingPipe is asuite of Java tools designedtoperform linguistic analysis on natural language data. The toolsare fast androbust enoughtobe used in acustomer-facingcommercial system. LingPipe's flexibility andincludedsource make it appropriate for research use. LingPipetoolsinclude astatistical named-entity detector,textclassificationand clustering.Inthisexperiment, we applythe LDA algorithmprovided underthe LingPipe API and set the number of topicsequal to 50 and thenumberofepochs to 2,000.

3 SOS Children's VillagesWeb site.http://www.soschildrensvillages.org.uk/charity-news/wikipedia-forschools.htm 4 LingPipe. http://alias-i.com/lingpipe

199 Figure 3: Examplesoftopics generated by using the LDA algorithm.

Figure 3shows someexamples of topics generated by theLDA algorithm. Each table lists thetop-10 terms rankedbythe probabilistic values. It can be observed that theLDA couldconceptually cluster highlysimilar termsintothe sametopics. Forexample,the termsart, gallery and paintingare assigned into thesametopic of 32.Onthe otherhand, thetopic 24 contains thetermsrelated to thebasic scientificelements and topic 27 contains thetermsrelated to sports.

We applied the article recommendationapproachdescribedinthe previoussection on a sample set of articles. Figure 4shows thecomparisonofthe links within thearticle and thelinks from recommendation. The bold text shows recommended article links that not foundinthe article linkmadebyhuman authors.

200 Figure 4: Examples of articlerecommendation based on thetopic-model approach.

201 The accuracy of theproposed recommendationapproach is evaluated by thehuman assessor. Thefive assessors receive the article title, the linked articles and the recommended articlesbyour Topic-Based model.The assessorassignedthe scoresfor each linked articles (LINK)and recommended articles (REC).The score is on the scale of 1to5.The average scores areshown in Table 2.

Table2:Evaluationresults betweenthe linkedarticles(LINK)and the recommended articles (REC)

Score Article LINKREC Bill Clinton2.275 3.675 Trigonometry2.375 3.725 Mona Lisa 2.43.8 Television2.125 3.125 Dinosaur 2.74.475 Cancer 2.0753.475 Average 2.3253.7125

Theresult shows that thescores from therecommended articles is higher than thescores fromlinkedarticles. This is especially truewhenthe articles are about the definition of something and many articles are the class or specific type of that article, e.g., there are many dinosaurtypearticles that related to dinosaur definitionarticle.

4Conclusion and futureworks

Wikipediaisawell-knownfree-content encyclopedia. Thecontent of Wikipediacan be navigated by browsing on apictorial subject indexoratitle word index of all topics. Organizing articles into the subject categoryset provides usersaconvenientway to access the articles on thesamesubject. Each article contains many hypertext links to otherarticleswhich are related to thecurrent article. However,the linkswhich were assignedbythe authors of thearticle cannot fully cover all related articles. Oneofthe reasons is due to the fact that thereisnoterm describing related articles within the current article.Inthispaper,weproposedatopic-model basedmethod for recommending related articlesinWikipedia Selectionfor Schools. Thetopicmodel is generated by usingthe LatentDirichlet Allocation(LDA) algorithm. The experimental results showedthatthe proposed method could help discover additionalrelated articles, someofwhich are not listedashyperlinks within agiven article.The proposed recommend articles improve relevance score by 59.68%.

202 Ourfutureworks include theconstruction of an evaluationcorpus.Aset of random articles will be selected and all related articles will be judged by human experts. The corpusisusefulinperforming theempirical analysis of adjusting theLDA parameters. In this paper, we constructed theLDA model fromtextual informationwithin thegiven articles.Inour next work,wewill extendthe LDA modelbyincluding theneighboring information surrounding thecurrentarticle. Theneighboring informationis, for example, theanchortexts of linksintothe currentarticle. Using theneighboring informationcould provide richerand more coverage of informationusedtodescribethe currentarticle.

References

[Bi08] Biro, I.; et al.: AComparative Analysis of LatentVariable Models for Web Page Classification.Latin American WebConference,pp.23-28,2008. [BNJ03]Blei, D. M.;Ng, A. Y.;Jordan, M. I.:Latentdirichletallocation. Journal of Machine Learning Research, 3(5): 993-1022, 2003. [Co07] Cosley,D.etal.: SuggestBot: Using IntelligentTaskRouting to Help People Find Work in Wikipedia: Proc. of the 12th acm international conferenceonintelligent user interfaces,New York, 2007. [De90]Deerwester S. et al.:Indexing by latentsemantic analysis. Journal of the American SocietyofInformation Science, 41(6):391-407, 1990. [FR05] FissahaAdafre,S.; Rijke, M.: Discovering missing links in wikipedia: Proc. of the 3rd int. workshop on link discovery, 2005, pp. 90-97. [GS04] Griffiths, T.; Steyvers,M.: Finding scientifictopics: Proc.ofthe NationalAcademy of Sciences, 2004,pp. 5228-5235. [HD08]Haruechaiyasak,C;Damrongrat C.: Article Recommendation Based on aTopic Model for Wikipedia Selection for Schools: Proc. of the11thInternational Conference on Asian Digital Libraries, 2008,S.339-342. [MC05] McCallum, A.;Corrada-Emmanuel, A.;Wang, X.: Topic and role discoveryinsocial networks:Proc. of IJCAI, 2005, pp. 786-791. [Ne06] Newman, D. et al.:Analyzing entities and topicsinnews articlesusing statistical topic models.InLecture NotesonComputer Science,Springer-Verlag, 2006. [SG06] Steyvers, M.; Griffiths, T.:Probabilistic topicmodels, In T.,Landauer, D.,McNamara, S., Dennis, and W.,Kintsch,(eds), LatentSemantic Analysis: ARoad to Meaning, Laurence Erlbaum, 2006. [Vo05] Voss, J.: Measuring Wikipedia:Proc. of Int. Conf.ofthe International Society for Scientometricsand Informetics, Stockholm, 2005.

203 Linking the tele-TASK video portal to the Semantic Web

Bert Baumann, Andreas Groß, Christoph Meinel, Harald Sack

Hasso Plattner Institute for Software Systems Engineering Postfach 900460, D-14440 Potsdam, Germany ([bert.baumann|andreas.gross|christoph.meinel|harald.sack]@hpi.uni-potsdam.de)

Abstract: Audiovisual data have gained an enormous and ever-growing popularity in the world wide web.Also agrowing number of educational content such as, e.g., lecture recordings or audiovisual learning material can be found recently.But, pinpoint and exhaustive retrievalofaudiovisual e-learning content in the web is rather difﬁcult as well as automated metadata interchange and integration. We demonstrate ause-case of metadata integration for audiovisual learning resources by complementing web pages of avideo lecture portal with semantic RDFaannotations giving waytoautomated access and universal retrievability.

1Introduction

Audiovisual data has become the predominant medium of the 20th century and the amount of video data available in the World Wide Web(WWW) is ever-growing day-to-day.Vi- deo portals and video search engines enable users to randomly access audiovisual data according to their demands and personal preferences. Among entertainment, news, or do- cumentaries there is also agrowing number of educational content available in the WWW. Numerous universities and institutions for higher education are publishing video recordings of lectures and seminars via streaming and podcasts, and some have become rather popular,such as, e.g., MIT open courseware or tele-TASK. The tele-TASK system offers an entire lifecycle infrastructure for lecture recording, archival, and retrieval. But, for the ordinary user,retrievaland access to those lecture recordings, is not always trivial. First, one has to know, where to find educational content. If one is looking for a specific lecture from MIT,then MIT’sweb site certainly is agood starting point. Without preferring aspecific educational institution, video portal’sorvideo search engines are the next best choice. But, neither special interest video portals nor video search engines provide exhaustive information about the universe of available lecture recordings and potential interrelationships, because metadata exchange formats for web based audiovisual learning resources are not utilized consistently.Furthermore, to be available for search engines, metadata have to be included into the web pages directly. Of course there exist metadata standards for documentation and e-learning content. Most prominent are Dublin Core (DC) metadata for bibliographical data and Learning Object Metadata (LOM) as well as the Sharable Content Object Reference Model (SCORM) for

204 the description of learning resources. Even though there are several ways for integrating one of these XML-based metadata directly into web pages, individual practice often inhi- bits real data integration of heterogeneous audiovisual learning resources. In this paper,weshowhow to use RDF-based semantic descriptions of DC and LOM, and howtointegrate this metadata directly into (X)HTML web pages via RDFa. There exist simple XSLTtransformations for extracting plain RDF metadata from RDFa-enriched web pages. We show, howtointegrate e-learning related metadata schemata with microformats and other RDF-based metadata (e.g., FOAF,DBPedia, etc. )toenable data integration overheterogeneous schemata. Thus, giving wayfor the development of newmashup applications and linking ownaudiovisual content to the Semantic Web’sLinked Open Data cloud. The paper is structured as follows: Section 2gives ashort overviewabout metadata schemata for bibliographical and audiovisual resources in the e-learning context or semantic metadata. Section 3introduces the tele-TASK video portal and lecture recording infrastructure, while Section 4provides implementational details about RDFaintegration of several metadata schemata into the tele-TASK web site and explains several examples on howtouse RDFa-based semantic metadata for e-learning resources. Section 5provides a short summary and outlook on future work.

2Metadata Standards and Semantics

2.1 Metadata

Metadata are data about data, i.e. structured data describing the characteristics of information bearing entities. Metadata can be used for identification, retrieval, evaluation, and administration of the data theydescribe. In particular there is an emphasis today on automated metadata processing with the purpose of identification and retrieval, as e.g. being applied in web search engines [Dur85]. Usually,metadata are categorized by their degree of inherent structure [DSS93]. On the lower end, there are unstructured metadata such as, free text annotations or tags. Structured metadata followadistinguished data schema. In addition, categories can be arranged by using taxonomic relationships for generalization and specialization. More complexmetadata structures comprise relationships, dependen- cies, constraints, and rules that can be expressed with the help of ontologies. Forefficient identification and retrieval, we have to followacommon, standardized metadata schema.

2.2 Metadata forDocumentation, Bibliography, and e-learning

Forthe purpose of documentation and bibliographyaswell as for e-learning, several metadata schemata have been developed. Well-established for bibliographyisthe so called Du- blin Core metadata standard, while for e-learning esp. Learning Object Metadata (LOM) has become popular.

205 Dublin Core The Dublic Core (DC) metadata standard wasdeveloped for the description of text-based information objects. It consists of 15 core elements that are intended for the compilation of bibliographical data [Wei97]. In addition, DCMI metadata terms recommend additional ﬁelds (element reﬁnements), which allowfor amore detailed description or categorization according to the user’spreferences. DC summarizes metadata for technical and content-based description of authors, related persons, intellectual property rights, as well as relationships among the described resources and life cycle information. Alt- hough intended to describe text-based resources, DC can be used to describe audiovisual objects such as, e.g., lecture recordings [HA99]. LOM Learning Object Metadata (LOM) is an open IEEE metadata standard for the description of learning objects [HD02]. The LOM metadata schema has been designed to support the reusability of learning objects, to aid discoverability,and to facilitate their interoperability,usually in the context of online learning management systems. It enables the description of entities related to the learning process such as, e.g., type of object, author,owner,terms of distribution, format, and pedagogical attributes, such as teaching or interaction style.

2.3 Semantic Metadata

The Semantic Webisanevolving extension of the World Wide Web(WWW) in which the meaning of information and services on the web is well defined [BLHL01]. Thereby, it will be possible for the web to understand and satisfy the requests of people and machines to use the web content. Keytechnology of the Semantic Webare semantic metadata (ontologies) representing commonly shared conceptualizations being specified with standardized, formalized languages [Gru93]. The World Wide WebConsortium (W3C)1 has already standardized aset of knowledge representation languages of different semantic expressivity being arranged in ahierarchically layered model. Resource Description Framework (RDF) und RDF Schema (RDFS) RDF and RDFS are simple knowledge representations for the definition of individual objects and their relationships as well as classes and their interrelationship among each other can be defined [LS99]. Individuals and concepts are identified via Uniform Resource Identifier (URI). RDF data consists out of simple triples (a,b,c),where a represents some individual, b stands for aproperty of a,and c givesadistinct value to property b.Individuals are concrete realizations (instances) of concepts (classes). Concepts can be derivedfrom more general concepts with the help of RDF Schema (RDFS) via generalization, specialization, or class extension. In that way, relationships among classes can be defined, also if theyare not part of the givenmetadata schema [BG04]. WebOntology Language (OWL) Semantic expressivity of RDF and RDFS is rather limited, as, e.g., there is no possibility to generalize statements for agroup of individuals, or the definition of logical attributes and constraints. The WebOntology Language OWLis the W3C standard for the specification of ontologies based on description logics [MvH04].

1http://w3c.org/

206

Semantic Web 2008-10-23

@prefix dc: . dc:title "Semantic Web" . dc:date "2008-10-23" .

Fig. 1: Example for RDFaand corresponding RDF extract (in RDF turtle syntax)

OWLcomes in three different variants, OWLLite, OWLDL, and OWLFull, according to its semantic expressivity,which is also related to its computational complexity.Inaddition to RDF(S) class and relationship deﬁnitions, OWLadds different class constructors, class and property constraints as well as (restricted) universal and existential quantiﬁcation.

2.4 Online Integration and Interoperability

While the metadata schemata that have been described in the previous sections are structured data, web pages are semi-structured (X)HTML-encoded information resources. (X)HTML only provides information about document structure and not about the document’stextual content. Also, (X)HTML cannot be extended to include other metadata. Nevertheless, there are different ways to incorporate additional metadata into (X)HTML-encoded documents. We will focus on microformats and RDFa: Microformats Microformats define aspecific markup format for semantic annotation of (X)HTML documents [Dub05]. Microformat annotations are encoded within (X)HTML tag attributes and can easily be extracted from web documents. Thus, applications are able to gather some information about the meaning of web page’scontent (such as contact information, geographic coordinates, calendar events, and the like) for subsequent processing. Microformat semantic is defined by common agreement and not by formal definition. RDFa Similar to microformats, RDFa(RDF in (X)HTML attributes) utilizes unused (X)HTML attributes to include RDF metadata into simple web pages [adi08]. RDFauses attributes from (X)HTML’s meta and link elements, and generalizes them so that theycan be used for all (X)HTML syntax elements (cf. Fig. 1). While microformats are always fixed to aspecial topic (calendar,geographic data, address data, etc.), RDFaannotations can makeuse of anyRDF ontology and thus, it is much more flexible and allows to annotate (X)HTML markup with semantics. Asimple mapping is defined with GRDDL (Gleaning Resource Descriptions from Dialects of Languages) [W3C07] and XSLT(Extensible Stylesheet Language Transformations) [Kay07] so that plain RDF may be extracted.

207 3Video lecturing with tele-TASK

This chapter introduces the tele-TASK system and its components, i.e. the tele-TASK recording system and the tele-TASK web portal.

3.1 The tele-TASK Recording and Distribution System

The tele-TASK2 recording system [SM02] is asophisticated technology for the creation and transmission of advanced video presentations via the internet. This state-of-the-art solution is outstanding for its simplicity and dependability.Inaddition to high-quality video and audio of the lecturer the system delivers asynchronous video feed of the lecturer’s computer screen without installing anyadditional software on the lecturer’scomputer.This ability is singular and separates tele-TASK from anycompetitive lecture recording devices. With the help of the tele-TASK technology users worldwide can access to teaching courses and presentations using live streams or archivedrecordings. The presentations are available via internet and can be downloaded on portable devices [WLM07] such as, e.g., PDA, mobile video players, 3G mobile phones, or lean-back consumer electronics also. The tele-TASK content is published via several distribution channels. The main distribution platform is the tele-TASK web portal (Fig. 2) for the publication of lecture and event recordings at the „Hasso Plattner Institute for Systems Engineering“ (HPI). In addition, podcasts of tele-TASK recordings can also be accessed via iTunes U3.Recorded tele- TASK lectures are available in various formats such as, e.g., RealMedia, Flash Video, and MP4. The portal offers different post-processing steps for cutting, synchronization, and media conversion. Currently the tele-TASK database comprises more than 2.200 lecture recordings and 2.600 podcasts from 500+ different speakers. All tele-TASK content can be accessed and downloaded for free. Since 2009 tele-TASK lectures are part of the popular Apple iTunes Urepository,which is part of the iTunes Store. But, as being part of the educational section of iTunes, tele- TASK content is freely available. The HPI’stele-TASK pool on iTunes Uisone out of four selected german elite education centers distributing their learning materials on iTunes U4. The iTunes store can only be used via the proprietary Apple iTunes client software. The main drawback of iTunes Ulies in its strictly proprietary nature, preventing worldwide searchability and data integration.

3.2 Restrictions of the tele-TASK Portal

Audiovisual lecture recordings in terms of streaming media are receiving an ever-growing popularity among learners. One of the reasons for it’spopularity is that the learner might

2(TeleTeaching Anywhere Solution Kit) 3http://itunes.hpi.uni-potsdam.de 4HPI iTunes Uportal page: http://deimos3.apple.com/WebObjects/Core.woa/Browse/hpi-de-public

208 Fig. 2: Reference of avideo lecture within the tele-TASK web portal

access and learn the video lecture everywhere at anytime, independent from the live event. Therefore, the retrievability of lecture recordings has become most decisive.Eventhough tele-TASK provides search formulas with automated online completion, it is difﬁcult for the user to specify the accurate search terms. In particular,the integration of additional external data resources such as, e.g., calendar,address books, or others, to increase the retrievability is not possible. One waytoachieve better search results and simplify the retrievalprocess is to provide semantic metadata to enable aﬂexible and dynamic data integration. Moreoverthe integration of semantic metadata enables the automatic connection e-learning resources worldwide.

4Integrating tele-Task into the Semantic Web

This chapter addresses the implementation of semantic metadata for the tele-TASK web portal via RDFaannotations and linking to the Semantic Web.

4.1 Standardized Metadata fortele-TASK Data and RDFaIntegration

The core data element of the tele-TASK database is the video recording of asingle lecture event. Lectures can be combined into groups of lectures or lecture series. Alecture series comprises all single lectures of adistinct topic within the time frame of asemester.Li- kewise, asingle lecture may be sectioned into several chapters, which are utilized for the production of podcast contributions. Metadata for lectures, such as, e.g., title, abstract, language, date, duration, lecturer name(s), etc., are complemented by lecture series metadata such as, e.g., keywords, series type, place, institution, etc. All tele-TASK metadata can be mapped to standardized XML- based metadata schemata such as DC and LOM (cf. Section 2). To be publicly available

209

Universelle Vokabularien mit XML (german)

01:32:35 Dr. Harald Sack ...

Die Vision des Semantic Web

...

Fig. 3: Dublin Core and LOM as Part of the (X)HTML WebPage on the WWW these metadata schemata must be integrated into (X)HTML encoded web pages. One waytoachieve this is to store metadata as aseparate XML-file being linked by the original web page. But, this approach prohibits proper assignment of single metadata to their corresponding representativesinthe (X)HTML file. Therefore, we decided to include metadata directly into the (X)HTML document via tag attributes and already available text information of the (X)HTML document. DC and LOM metadata schemata are already available as RDF encoded metadata [NPB03] and can be directly included into (X)HTML documents via RDFa. Forfurther processing, RDF syntax can be extracted automatically via XSLTorbyusing applications such as, e.g., W3C RDF extractor5. By including DC and LOM RDF Schema definitions via namespaces into the (X)HTML document, these metadata schemata can be utilized with RDFaannotations. One wayto supply data values can be achievedbyintegrating the already displayed textual content of the (X)HTML web page into the RDFaannotation. RDF tripels represent assertions about resources (identified via URI) in the same wayassimple natural language statements consisting of subject (resource a), predicate (relationship or property b)and an object value (literal or resource c). ViaRDFaannotation subject a and property b can be embedded as attributes within an (X)HTML-tag that envelopes atextvalue, the object value c,i.e.

. If aRDF triple is about to contain data values that are not displayed in the (X)HTML web page, the content-attribute can be used to hold the data, i.e.

. Fig. 3shows atypical example of atele-TASK (X)HTML web page withembedded RDFa annotation displaying metadata about video lectures.

5http://www.w3.org/2007/08/pyRdfa/extract

210 PREFIX foaf: SELECT ?title ?lecture WHERE { ?lecture dc:title ?title FILTER regex(str(?title), "rdf", "i"). }

Fig. 4: SPARQL example query to search all lectures that contain the string »rdf« in the title

Fig. 5: Sesame server web portal displaying the result of the query in Fig. 4

4.2 Using RDFaAnnotation forInformation Integration in the Semantic Web

By providing semantic metadata via RDF/RDFaannotation the tele-TASK data can be combined and complemented with alarge variety external data sources. On the one hand, autonomous software agents, search engines, or applications can link their ownresources with tele-TASK data by using aSPARQL endpoint, while on the other hand, tele-TASK data also can be connected and augmented with external semantic data being matched with tele-TASK metadata. naddition to the RDFaannotation being included within the tele-TASK (X)HTML web pages, we decided to store RDF data also in the RDF triple store database Sesame6.Sesame is an open source RDF database with support for RDF Schema inferencing and querying providing also aSPARQL endpoint. ASPARQL endpoint enables users (human as well as applications) to query aRDF knowledge base via the SPARQL language. ASPARQL endpoint typically returns query results in various machine-processable formats. Fig. 4 shows asimple example query in SPARQL and Fig. 5showthe result being displayed in the Sesame web user interface. By providing aSPARQL endpoint and by deploying RDF-based Dublin Core and LOM metadata, tele-TASK video resources are linked to the semantic web, i.e. theyprovide a meaningful interface that can be accessed by human users as well as by software applica-

6http://www.openrdf.org/

211

Fig. 6: Personal information with vCard and FOAF tions. Applications can automatically search for authors, titles, or media formats, and can evaluate additional metadata provided for each video lecture or video lecture series, such as, e.g., keywords, duration, date, etc. Authors, as being persons, can also be described with the help of alternative metadata schemata for personal and address information, such as vCard7 or FOAF 8 .FOAFdefines aset of terms for letting users describe persons, their activities and their relations to other people and objects [BM07]. Anyone can use FOAF to describe himself or herself. In difference to other social networking services, FOAF allows groups of people to describe social networks without the need for acentralized database. FOAF is one of the largest projects on the Semantic Webwhich has an estimated 2–5 million users. vCard is an electronic format for the consistently exchange of business information. vCard elements can be freely reused, as e.g., within different LOM-attributes as being shown in fig. fig:vCard, which givesanexample of vCard and FOAF data integration via RDFaannotation.

5Summary and Outlook

We have shown ause-case of semantic data Integration via RDFabycomplementing the tele-TASK web portal with semantic metadata that is already available within the ordinary tele-TASK database by deploying Dublin Core and LOM metadata schemata for RDFa data integration. As anextstep, the SPARQL endpoint being described in the previous section will also be opened up for the public. This enables tele-TASK to link with the worldwide semantic web.Uptonow,only the video data being provided by the tele-TASK web portal are described via standard metadata schemata. But, to be able to access the meaning of its content, domain ontologies must be deployed to annotate also the video content. Aﬁrst step will be the mapping of already existing keywords that describe the video data’scontent to concepts, classes, and instances of the global wikipedia encyclo-

7http://www.w3.org/TR/vcard-rdf 8Friend of aFriend (FOAF) project homepage, http://www.foaf-project.org/

212 pedia, in particular to it’ssemantic counterpart, the DBPedia9.This would be the ﬁrst step for tele-TASK to participate in the Linked Open Data (LOD) community10. Furthermore, if tele-Task entities can be mapped to LOD entities such as, e.g. DBPedia entities, numerous additional information stemming from the popular online encyclopedia ’Wikipedia’11 can be used to complement and to enrich tele-Task data on the tele-Task portal site as well as for iTunes U. E.G., information about famous speakers and presenters, who are also present at wikipedia, can be extracted in an automated wayand presented on the tele-Task website for that speaker.

Acknowledgment

The authors would liketothank students of the seminar „Semantic Webenabled Software“ of winter semester 2008/09, Andreas Motl, Paul Schröder and Dennis Brutski.

References

[adi08] RDFainXHTML: Syntax and Processing, W3C Recommendation, 2008.

[ADL04] Advanced Distributed Learning ADL. Sharable Content Object Reference Model (SCORM) 2nd Edition Conformance Requirements Version 1.1.2004.

[BG04] Dan Brickleyund R.V.Guha. Resource Description Framework (RDF) Schema Speciﬁ- cation, February 2004.

[BLHL01] TimBerners-Lee, James Hendler und Ora Lassila. The Semantic Web. Scientiﬁc Ame- rican,284(5):34–43, 2001.

[BM07] Dan Brickleyund Libby Miller.The Friend Of AFriend (FOAF) Vocabulary Speciﬁcati- on, November 2007.

[DSS93] Randall Davis, Howard Shrobe und Peter Szolovits. What is aKnowledge Representation. AI Magazine,14(1):17–33, 1993.

[Dub05] Micah Dubinko. What are Microformats, March 2005.

[Dur85] William R. Durrell. Data Administration: APractical Guide to Data Administration. McGraw-Hill, 1985.

[Gru93] Thomas R. Gruber.ATranslation Approach to Portable Ontology Speciﬁcations. Know- ledgeAcquisitation,5(2):199–220, 1993.

[HA99] Jane Hunter und Liz Armstrong. Acomparison of schemas for video metadata representation. Comput. Netw.,31(11-16):1431–1451, 1999.

9http://dbpedia.org/ 10http://linkeddata.org 11http://www.wikipedia.org/

213 [HD02] Wayne Hodgins und Erik Duval. Draft standard for learning technology -Learning Object Metadata -ISO/IEC 11404. Bericht, 2002.

[Kay07] Michael Kay.XSL Transformations (XSLT) Version 2.0. W3C recommendation, W3C, January 2007. http://www.w3.org/TR/2007/REC-xslt20-20070123/.

[LS99] O. Lassila und R. R. Swick. Resource Description Framework (RDF) Model and Syntax Speciﬁcation .W3C Recommendation, World Wide WebConsortium, 1999.

[MvH04] Deborah L. McGuinness und Frank vanHarmelen. OWLWeb Ontology Language: Over- view, W3C Recommendation, 10 February 2004.

[NPB03] Mikael Nilsson, Matthias Palmer und Jan Brase. The LOM RDF binding -principles and implementation. In Proc. of 3rdAnnual Ariadne Conference,2003.

[SM02] V. Schillings und Ch. Meinel. Tele-TASK –tele-teaching anywhere solution kit. In Proceedings of ACMSIGUCCS,Providence, USA, 2002.

[W3C07] W3C. Gleaning Resource Descriptions from Dialects of Languages (GRDDL). W3c recommendation, W3C, September 2007.

[Wei97] Stuart Weibel. The Dublin Core: ASimple Content Description Model for Electronic Resources, 1997.

[WLM07] Katrin Wolf, Serge Linckels und Christoph Meinel. Teleteaching anywhere solution kit (Tele-TASK) goes mobile. In SIGUCCS ’07: Proceedings of the 35th annual ACM SIGUCCS conference on User services,Seiten 366–371, NewYork, NY,USA, 2007. ACM.

214

Index of Authors

Bohme,¨ Thomas ...... 149 Sack, Harald ...... 204 Bakalov, Fedor ...... 123 Sanmateu, Maria-Amparo ...... 27 Baumann, Bert ...... 204 Schau, Volkmar ...... 97 Schreyer,Jens ...... 149 Cai, Wentong ...... 157 Send, Hendrik ...... 36 Sriurai, Wongkot ...... 194 Dobes, Zuzana Krifka ...... 49 Strecker,Conny...... 183 Dressler,Enrico ...... 70 Strecker,Thomas ...... 183 Eichler,Gerald ...... 183 Svagard,˚ Ingrid ...... 60 Eisemann, Matthias ...... 109 Tang, Wallace ...... 85 Erfurth, Christian ...... 97 Tavangarian, Djamshid ...... 70 Groß, Andreas ...... 204 Telschow, Anke...... 109 Trier,Matthias ...... 27 Haruechaiyasak, Choochart ...... 194 Unger,Herwig ...... 169 Konig-Ries,¨ Birgitta ...... 123 Kirchner,Kathrin ...... 15 Vent, Kathrin ...... 135

Luke,¨ Karl-Heinz ...... 49, 109 Welsch, Martin ...... 123 Lertsuwanakul, Lada-On ...... 169 Wetzker,Robert ...... 183 Lienicke, Andreas ...... 27 Wienhofen, Leendert ...... 60 Liu, Cheng ...... 157 Lommatzsch, Andreas ...... 183 Zender,Raphael ...... 70 Lucke, Ulrike...... 70

Mugge,¨ Holger ...... 109 Meesad, Phayung ...... 194 Meinel, Christoph ...... 204 Michelis, Daniel ...... 36

Nabeth, Thierry ...... 15 Nauerz, Andreas ...... 123 Ng, Ka-ho ...... 85

Pasold, Rene.´ ...... 97 Ploch, Danuta ...... 183

Razmerita, Liana ...... 15 Rederer,Andreas ...... 27, 49 Rossak, Wilhelm ...... 97

217

GI-Edition Lecture Notes in Informatics

P-1 Gregor Engels, Andreas Oberweis, Albert P-18 Elmar J. Sinz, Markus Plaha (Hrsg.): Zündorf (Hrsg.): Modellierung 2001. Modellierung betrieblicher Informations- P-2 Mikhail Godlevsky, Heinrich C. Mayr systeme – MobIS 2002. (Hrsg.): Information Systems Technology P-19 Sigrid Schubert, Bernd Reusch, Norbert and its Applications, ISTA’2001. Jesse (Hrsg.): Informatik bewegt – Infor- P-3 Ana M. Moreno, Reind P. van de Riet matik 2002 – 32. Jahrestagung der Gesell- (Hrsg.): Applications of Natural Lan- schaft für Informatik e.V.(GI) 30.Sept.- guage to Information Systems, 3.Okt. 2002 in Dortmund. NLDB’2001. P-20 Sigrid Schubert, Bernd Reusch, Norbert P-4 H. Wörn, J. Mühling, C. Vahl, H.-P. Jesse (Hrsg.): Informatik bewegt – Infor- Meinzer (Hrsg.): Rechner- und sensor- matik 2002 – 32. Jahrestagung der Gesell- gestützte Chirurgie; Workshop des SFB schaft für Informatik e.V.(GI) 30.Sept.- 414. 3.Okt. 2002 in Dortmund (Ergänzungs- band). P-5 Andy Schürr (Hg.): OMER – Object- Oriented Modeling of Embedded Real- P-21 Jörg Desel, Mathias Weske (Hrsg.): Time Systems. Promise 2002: Prozessorientierte Metho- den und Werkzeuge für die Entwicklung P-6 Hans-Jürgen Appelrath, Rolf Beyer, Uwe von Informationssystemen. Marquardt, Heinrich C. Mayr, Claudia Steinberger (Hrsg.): Unternehmen Hoch- P-22 Sigrid Schubert, Johannes Magenheim, schule, UH’2001. Peter Hubwieser, Torsten Brinda (Hrsg.): Forschungsbeiträge zur “Didaktik der P-7 Andy Evans, Robert France, Ana Moreira, Informatik” – Theorie, Praxis, Evaluation. Bernhard Rumpe (Hrsg.): Practical UML- Based Rigorous Development Methods – P-23 Thorsten Spitta, Jens Borchers, Harry M. Countering or Integrating the extremists, Sneed (Hrsg.): Software Management pUML’2001. 2002 – Fortschritt durch Beständigkeit P-8 Reinhard Keil-Slawik, Johannes Magen- P-24 Rainer Eckstein, Robert Tolksdorf heim (Hrsg.): Informatikunterricht und (Hrsg.): XMIDX 2003 – XML- Medienbildung, INFOS’2001. Technologien für Middleware – Middle- ware für XML-Anwendungen P-9 Jan von Knop, Wilhelm Haverkamp (Hrsg.): Innovative Anwendungen in P-25 Key Pousttchi, Klaus Turowski (Hrsg.): Kommunikationsnetzen, 15. DFN Arbeits- Mobile Commerce – Anwendungen und tagung. Perspektiven – 3. Workshop Mobile Commerce, Universität Augsburg, P-10 Mirjam Minor, Steffen Staab (Hrsg.): 1st 04.02.2003 German Workshop on Experience Man- agement: Sharing Experiences about the P-26 Gerhard Weikum, Harald Schöning, Sharing Experience. Erhard Rahm (Hrsg.): BTW 2003: Daten- banksysteme für Business, Technologie P-11 Michael Weber, Frank Kargl (Hrsg.): und Web Mobile Ad-Hoc Netzwerke, WMAN 2002. P-27 Michael Kroll, Hans-Gerd Lipinski, Kay Melzer (Hrsg.): Mobiles Computing in P-12 Martin Glinz, Günther Müller-Luschnat der Medizin (Hrsg.): Modellierung 2002. P-28 Ulrich Reimer, Andreas Abecker, Steffen P-13 Jan von Knop, Peter Schirmbacher and Staab, Gerd Stumme (Hrsg.): WM 2003: Viljan Mahni_ (Hrsg.): The Changing Professionelles Wissensmanagement – Er- Universities – The Role of Technology. fahrungen und Visionen P-14 Robert Tolksdorf, Rainer Eckstein P-29 Antje Düsterhöft, Bernhard Thalheim (Hrsg.): XML-Technologien für das Se- (Eds.): NLDB’2003: Natural Language mantic Web – XSW 2002. Processing and Information Systems P-15 Hans-Bernd Bludau, Andreas Koop P-30 Mikhail Godlevsky, Stephen Liddle, (Hrsg.): Mobile Computing in Medicine. Heinrich C. Mayr (Eds.): Information P-16 J. Felix Hampe, Gerhard Schwabe (Hrsg.): Systems Technology and its Applications Mobile and Collaborative Busi-ness 2002. P-31 Arslan Brömme, Christoph Busch (Eds.): P-17 Jan von Knop, Wilhelm Haverkamp BIOSIG 2003: Biometric and Electronic (Hrsg.): Zukunft der Netze –Die Verletz- Signatures barkeit meistern, 16. DFN Arbeitstagung. P-32 Peter Hubwieser (Hrsg.): Informatische P-48 Anatoly Doroshenko, Terry Halpin, Fachkonzepte im Unterricht – INFOS Stephen W. Liddle, Heinrich C. Mayr 2003 (Hrsg.): Information Systems Technology P-33 Andreas Geyer-Schulz, Alfred Taudes and its Applications (Hrsg.): Informationswirtschaft: Ein P-49 G. Schiefer, P. Wagner, M. Morgenstern, Sektor mit Zukunft U. Rickert (Hrsg.): Integration und Daten- P-34 Klaus Dittrich, Wolfgang König, Andreas sicherheit – Anforderungen, Konflikte und Oberweis, Kai Rannenberg, Wolfgang Perspektiven Wahlster (Hrsg.): Informatik 2003 – P-50 Peter Dadam, Manfred Reichert (Hrsg.): Innovative Informatikanwendungen INFORMATIK 2004 – Informatik ver- (Band 1) bindet (Band 1) Beiträge der 34. Jahresta- P-35 Klaus Dittrich, Wolfgang König, Andreas gung der Gesellschaft für Informatik e.V. Oberweis, Kai Rannenberg, Wolfgang (GI), 20.-24. September 2004 in Ulm Wahlster (Hrsg.): Informatik 2003 – P-51 Peter Dadam, Manfred Reichert (Hrsg.): Innovative Informatikanwendungen INFORMATIK 2004 – Informatik ver- (Band 2) bindet (Band 2) Beiträge der 34. Jahresta- P-36 Rüdiger Grimm, Hubert B. Keller, Kai gung der Gesellschaft für Informatik e.V. Rannenberg (Hrsg.): Informatik 2003 – (GI), 20.-24. September 2004 in Ulm Mit Sicherheit Informatik P-52 Gregor Engels, Silke Seehusen (Hrsg.): P-37 Arndt Bode, Jörg Desel, Sabine Rath- DELFI 2004 – Tagungsband der 2. e- mayer, Martin Wessner (Hrsg.): DeLFI Learning Fachtagung Informatik 2003: e-Learning Fachtagung Informatik P-53 Robert Giegerich, Jens Stoye (Hrsg.): P-38 E.J. Sinz, M. Plaha, P. Neckel (Hrsg.): German Conference on Bioinformatics – Modellierung betrieblicher Informations- GCB 2004 systeme – MobIS 2003 P-54 Jens Borchers, Ralf Kneuper (Hrsg.): P-39 Jens Nedon, Sandra Frings, Oliver Göbel Softwaremanagement 2004 – Outsourcing (Hrsg.): IT-Incident Management & IT- und Integration Forensics – IMF 2003 P-55 Jan von Knop, Wilhelm Haverkamp, Eike P-40 Michael Rebstock (Hrsg.): Modellierung Jessen (Hrsg.): E-Science und Grid Ad- betrieblicher Informationssysteme – Mo- hoc-Netze Medienintegration bIS 2004 P-56 Fernand Feltz, Andreas Oberweis, Benoit P-41 Uwe Brinkschulte, Jürgen Becker, Diet- Otjacques (Hrsg.): EMISA 2004 – Infor- mar Fey, Karl-Erwin Großpietsch, Chris- mationssysteme im E-Business und E- tian Hochberger, Erik Maehle, Thomas Government Runkler (Edts.): ARCS 2004 – Organic P-57 Klaus Turowski (Hrsg.): Architekturen, and Pervasive Computing Komponenten, Anwendungen P-42 Key Pousttchi, Klaus Turowski (Hrsg.): P-58 Sami Beydeda, Volker Gruhn, Johannes Mobile Economy – Transaktionen und Mayer, Ralf Reussner, Franz Schweiggert Prozesse, Anwendungen und Dienste (Hrsg.): Testing of Component-Based P-43 Birgitta König-Ries, Michael Klein, Systems and Software Quality Philipp Obreiter (Hrsg.): Persistance, P-59 J. Felix Hampe, Franz Lehner, Key Scalability, Transactions – Database Me- Pousttchi, Kai Ranneberg, Klaus Turowski chanisms for Mobile Applications (Hrsg.): Mobile Business – Processes, P-44 Jan von Knop, Wilhelm Haverkamp, Eike Platforms, Payments Jessen (Hrsg.): Security, E-Learning. E- P-60 Steffen Friedrich (Hrsg.): Unterrichtskon- Services zepte für inforrmatische Bildung P-45 Bernhard Rumpe, Wofgang Hesse (Hrsg.): P-61 Paul Müller, Reinhard Gotzhein, Jens B. Modellierung 2004 Schmitt (Hrsg.): Kommunikation in ver- P-46 Ulrich Flegel, Michael Meier (Hrsg.): teilten Systemen Detection of Intrusions of Malware & P-62 Federrath, Hannes (Hrsg.): „Sicherheit Vulnerability Assessment 2005“ – Sicherheit – Schutz und Zuver- P-47 Alexander Prosser, Robert Krimmer lässigkeit (Hrsg.): Electronic Voting in Europe – P-63 Roland Kaschek, Heinrich C. Mayr, Technology, Law, Politics and Society Stephen Liddle (Hrsg.): Information Sys- tems – Technology and ist Applications P-64 Peter Liggesmeyer, Klaus Pohl, Michael P-80 Mareike Schoop, Christian Huemer, Goedicke (Hrsg.): Software Engineering Michael Rebstock, Martin Bichler 2005 (Hrsg.): Service-Oriented Electronic P-65 Gottfried Vossen, Frank Leymann, Peter Commerce Lockemann, Wolffried Stucky (Hrsg.): P-81 Wolfgang Karl, Jürgen Becker, Karl- Datenbanksysteme in Business, Techno- Erwin Großpietsch, Christian Hochberger, logie und Web Erik Maehle (Hrsg.): ARCS´06 P-66 Jörg M. Haake, Ulrike Lucke, Djamshid P-82 Heinrich C. Mayr, Ruth Breu (Hrsg.): Tavangarian (Hrsg.): DeLFI 2005: 3. Modellierung 2006 deutsche e-Learning Fachtagung Infor- P-83 Daniel Huson, Oliver Kohlbacher, Andrei matik Lupas, Kay Nieselt and Andreas Zell P-67 Armin B. Cremers, Rainer Manthey, Peter (eds.): German Conference on Bioinfor- Martini, Volker Steinhage (Hrsg.): matics INFORMATIK 2005 – Informatik LIVE P-84 Dimitris Karagiannis, Heinrich C. Mayr, (Band 1) (Hrsg.): Information Systems Technology P-68 Armin B. Cremers, Rainer Manthey, Peter and its Applications Martini, Volker Steinhage (Hrsg.): P-85 Witold Abramowicz, Heinrich C. Mayr, INFORMATIK 2005 – Informatik LIVE (Hrsg.): Business Information Systems (Band 2) P-86 Robert Krimmer (Ed.): Electronic Voting P-69 Robert Hirschfeld, Ryszard Kowalcyk, 2006 Andreas Polze, Matthias Weske (Hrsg.): NODe 2005, GSEM 2005 P-87 Max Mühlhäuser, Guido Rößling, Ralf Steinmetz (Hrsg.): DELFI 2006: 4. e- P-70 Klaus Turowski, Johannes-Maria Zaha Learning Fachtagung Informatik (Hrsg.): Component-oriented Enterprise Application (COAE 2005) P-88 Robert Hirschfeld,Andreas Polze, Ryszard Kowalczyk (Hrsg.): NODe 2006, P-71 AndrewTorda, Stefan Kurz, Matthias GSEM 2006 Rarey (Hrsg.): German Conference on Bioinformatics 2005 P-90 Joachim Schelp, Robert Winter, Ulrich Frank, Bodo Rieger, Klaus Turowski P-72 Klaus P. Jantke, Klaus-Peter Fähnrich, (Hrsg.): Integration, Informationslogistik Wolfgang S. Wittig (Hrsg.): Marktplatz und Architektur Internet: Von e-Learning bis e-Payment P-91 Henrik Stormer, Andreas Meier, Michael P-73 Jan von Knop, Wilhelm Haverkamp, Eike Schumacher (Eds.): European Conference Jessen (Hrsg.): “Heute schon das Morgen on eHealth 2006 sehen“ P-92 Fernand Feltz, Benoît Otjacques, Andreas P-74 Christopher Wolf, Stefan Lucks, Po-Wah Oberweis, Nicolas Poussing (Eds.): AIM Ya u (Hrsg.): WEWoRC 2005 – Western 2006 European Workshop on Research in Cryptology P-93 Christian Hochberger, Rüdiger Liskowsky (Eds.): INFORMATIK 2006 – Informatik P-75 Jörg Desel, Ulrich Frank (Hrsg.): Enter- für Menschen, Band 1 prise Modelling and Information Systems Architecture P-94 Christian Hochberger, Rüdiger Liskowsky (Eds.): INFORMATIK 2006 – Informatik P-76 Thomas Kirste, Birgitta König-Riess, Key für Menschen, Band 2 Pousttchi, Klaus Turowski (Hrsg.): Mo- bile Informationssysteme – Potentiale, P-95 Matthias Weske, Markus Nüttgens (Eds.): Hindernisse, Einsatz EMISA 2005: Methoden, Konzepte und Technologien für die Entwicklung von P-77 Jana Dittmann (Hrsg.): SICHERHEIT dienstbasierten Informationssystemen 2006 P-96 Saartje Brockmans, Jürgen Jung, Yo rk P-78 K.-O.Wenkel, P. Wagner, M. Morgens- Sure (Eds.): Meta-Modelling and Ontolo- tern, K. Luzi, P. Eisermann (Hrsg.): Land- gies und Ernährungswirtschaft im Wandel P-97 Oliver Göbel, Dirk Schadt, Sandra Frings, P-79 Bettina Biel, Matthias Book, Volker Hardo Hase, Detlef Günther, Jens Nedon Gruhn (Hrsg.): Softwareengineering 2006 (Eds.): IT-Incident Mangament & IT- Forensics – IMF 2006 P-98 Hans Brandt-Pook, Werner Simonsmeier P-112 Sigrid Schubert (Hrsg.) und Thorsten Spitta (Hrsg.): Beratung in Didaktik der Informatik in der Softwareentwicklung – Modelle, Theorie und Praxis Methoden, Best Practices P-113 Sören Auer, Christian Bizer, Claudia P-99 Andreas Schwill, Carsten Schulte, Marco Müller,Anna V. Zhdanova (Eds.) Thomas (Hrsg.): Didaktik der Informatik The Social Semantic Web 2007 st P-100 Peter Forbrig, Günter Siegel, Markus Proceedings of the 1 Conference on Schneider (Hrsg.): HDI 2006: Hochschul- Social Semantic Web (CSSW) didaktik der Informatik P-114 Sandra Frings, Oliver Göbel, Detlef Günther, P-101 Stefan Böttinger, Ludwig Theuvsen, Hardo G. Hase, Jens Nedon, Dirk Schadt, Susanne Rank, Marlies Morgenstern (Hrsg.): Arslan Brömme (Eds.) Agrarinformatik im Spannungsfeld IMF2007 IT-incident zwischen Regionalisierung und globalen management & IT-forensics rd Wertschöpfungsketten Proceedings of the 3 International Conference on IT-Incident Management P-102 Otto Spaniol (Eds.): Mobile Services and & IT-Forensics Personalized Environments P-115 Claudia Falter,Alexander Schliep, P-103 Alfons Kemper, Harald Schöning, Thomas Joachim Selbig, Martin Vingron and Rose, Matthias Jarke, Thomas Seidl, Dirk Walther (Eds.) Christoph Quix, Christoph Brochhaus German conference on bioinformatics (Hrsg.): Datenbanksysteme in Business, GCB 2007 Technologie und Web (BTW 2007) P-116 Witold Abramowicz, Leszek Maciszek (Eds.) P-104 Birgitta König-Ries, Franz Lehner, Business Process and Services Computing Rainer Malaka, Can Türker (Hrsg.) 1st International Working Conference on MMS 2007: Mobilität und mobile Business Process and Services Computing Informationssysteme BPSC 2007 P-105 Wolf-Gideon Bleek, Jörg Raasch, P-117 Ryszard Kowalczyk (Ed.) Heinz Züllighoven (Hrsg.) Grid service engineering and manegement Software Engineering 2007 The 4th International Conference on Grid P-106 Wolf-Gideon Bleek, Henning Schwentner, Service Engineering and Management Heinz Züllighoven (Hrsg.) GSEM 2007 Software Engineering 2007 – P-118 Andreas Hein, Wilfried Thoben, Hans- Beiträge zu den Workshops Jürgen Appelrath, Peter Jensch (Eds.) P-107 Heinrich C. Mayr, European Conference on ehealth 2007 Dimitris Karagiannis (eds.) P-119 Manfred Reichert, Stefan Strecker, Klaus Information Systems Turowski (Eds.) Technology and its Applications Enterprise Modelling and Information P-108 Arslan Brömme, Christoph Busch, Systems Architectures Detlef Hühnlein (eds.) Concepts and Applications BIOSIG 2007: P-120 Adam Pawlak, Kurt Sandkuhl, Biometrics and Wojciech Cholewa, Electronic Signatures Leandro Soares Indrusiak (Eds.) P-109 Rainer Koschke, Otthein Herzog, Karl- Coordination of Collaborative Heinz Rödiger, Marc Ronthaler (Hrsg.) Engineering - State of the Art and Future INFORMATIK 2007 Challenges Informatik trifft Logistik P-121 Korbinian Herrmann, Bernd Bruegge (Hrsg.) Band 1 Software Engineering 2008 P-110 Rainer Koschke, Otthein Herzog, Karl- Fachtagung des GI-Fachbereichs Heinz Rödiger, Marc Ronthaler (Hrsg.) Softwaretechnik INFORMATIK 2007 P-122 Walid Maalej, Bernd Bruegge (Hrsg.) Informatik trifft Logistik Software Engineering 2008 - Band 2 Workshopband P-111Christian Eibl, Johannes Magenheim, Fachtagung des GI-Fachbereichs Sigrid Schubert, Martin Wessner (Hrsg.) Softwaretechnik DeLFI 2007: 5. e-Learning Fachtagung Informatik P-123 Michael H. Breitner, Martin Breunig, Elgar P-133 Heinz-Gerd Hegering, Axel Lehmann, Fleisch, Ley Pousttchi, Klaus Turowski (Hrsg.) Hans Jürgen Ohlbach, Christian Mobile und Ubiquitäre Scheideler (Hrsg.) Informationssysteme –Technologien, INFORMATIK 2008 Prozesse, Marktfähigkeit Beherrschbare Systeme –dank Informatik Proceedings zur 3. Konferenz Mobile und Band 1 Ubiquitäre Informationssysteme P-134 Heinz-Gerd Hegering, Axel Lehmann, (MMS 2008) Hans Jürgen Ohlbach, Christian P-124 Wolfgang E. Nagel, Rolf Hoffmann, Scheideler (Hrsg.) Andreas Koch (Eds.) INFORMATIK 2008 9th Workshop on Parallel Systems and Beherrschbare Systeme –dank Informatik Algorithms (PASA) Band 2 Workshop of the GI/ITG Speciel Interest P-135 Torsten Brinda, Michael Fothe, Groups PARS and PARVA Peter Hubwieser, Kirsten Schlüter (Hrsg.) P-125 Rolf A.E. Müller, Hans-H. Sundermeier, Didaktik der Informatik – Ludwig Theuvsen, Stephanie Schütze, Aktuelle Forschungsergebnisse Marlies Morgenstern (Hrsg.) P-136 Andreas Beyer, Michael Schroeder (Eds.) Unternehmens-IT: German Conference on Bioinformatics Führungsinstrument oder GCB 2008 Verwaltungsbürde Referate der 28. GIL Jahrestagung P-137 Arslan Brömme, Christoph Busch, Detlef Hühnlein (Eds.) P-126 Rainer Gimnich, Uwe Kaiser, Jochen BIOSIG 2008: Biometrics and Electronic Quante, Andreas Winter (Hrsg.) Signatures 10th Workshop Software Reengineering (WSR 2008) P-138 Barbara Dinter, Robert Winter, Peter Chamoni, Norbert Gronau, Klaus P-127 Thomas Kühne, Wolfgang Reisig, Turowski (Hrsg.) Friedrich Steimann (Hrsg.) Synergien durch Integration und Modellierung 2008 Informationslogistik P-128 Ammar Alkassar, Jörg Siekmann (Hrsg.) Proceedings zur DW2008 Sicherheit 2008 P-139 Georg Herzwurm, Martin Mikusz (Hrsg.) Sicherheit, Schutz und Zuverlässigkeit Industrialisierung des Software- Beiträge der 4. Jahrestagung des Managements Fachbereichs Sicherheit der Gesellschaft Fachtagung des GI-Fachausschusses für Informatik e.V.(GI) Management der Anwendungsentwick- 2.-4. April 2008 lung und -wartung im Fachbereich Saarbrücken, Germany Wirtschaftsinformatik P-129 Wolfgang Hesse, Andreas Oberweis (Eds.) P-140 Oliver Göbel, Sandra Frings, Detlef Sigsand-Europe 2008 Günther, Jens Nedon, Dirk Schadt (Eds.) Proceedings of the Third AIS SIGSAND IMF 2008 - IT Incident Management & European Symposium on Analysis, IT Forensics Design, Use and Societal Impact of Information Systems P-141 Peter Loos, Markus Nüttgens, Klaus Turowski, Dirk Werth (Hrsg.) P-130 Paul Müller, Bernhard Neumair, Modellierung betrieblicher Informations- Gabi Dreo Rodosek (Hrsg.) systeme (MobIS 2008) 1. DFN-Forum Kommunikations- Modellierung zwischen SOA und technologien Beiträge der Fachtagung Compliance Management P-131 Robert Krimmer, Rüdiger Grimm (Eds.) rd P-142 R. Bill, P. Korduan, L. Theuvsen, 3 International Conference on Electronic M. Morgenstern (Hrsg.) Voting 2008 Anforderungen an die Agrarinformatik Co-organized by Council of Europe, durch Globalisierung und Gesellschaft für Informatik and E- Klimaveränderung Voting.CC P-143 Peter Liggesmeyer, Gregor Engels, P-132 Silke Seehusen, Ulrike Lucke, Jürgen Münch, Jörg Dörr, Stefan Fischer (Hrsg.) Norman Riegel (Hrsg.) DeLFI 2008: Software Engineering 2009 Die 6. e-Learning Fachtagung Informatik Fachtagung des GI-Fachbereichs Softwaretechnik P-144 Johann-Christoph Freytag, Thomas Ruf, Wolfgang Lehner, Gottfried Vossen (Hrsg.) Datenbanksysteme in Business, Technologie und Web (BTW) P-145 Knut Hinkelmann, Holger Wache (Eds.) WM2009: 5th Conference on Professional Knowledge Management P-146 Markus Bick, Martin Breunig, Hagen Höpfner (Hrsg.) Mobile und Ubiquitäre Informationssysteme – Entwicklung, Implementierung und Anwendung 4. Konferenz Mobile und Ubiquitäre Informationssysteme (MMS 2009) P-147 Witold Abramowicz, Leszek Maciaszek, Ryszard Kowalczyk, Andreas Speck (Eds.) Business Process, Services Computing and Intelligent Service Management BPSC 2009 · ISM 2009 · YRW-MBP 2009 P-148 Christian Erfurth, Gerald Eichler, Volkmar Schau (Eds.) 9th International Conference on Innovative Internet Community Systems I2CS 2009 P-149Paul Müller, Bernhard Neumair, Gabi Dreo Rodosek (Hrsg.) 2. DFN-Forum Kommunikationstechnologien Beiträge der Fachtagung

The titles can be purchased at: Köllen Druck + Verlag GmbH Ernst-Robert-Curtius-Str. 14 · D-53117 Bonn Fax: +49 (0)228/9898222 E-Mail: [email protected]