@t or ead an d Cool: g in g f or Tim e, Task an d Em ot ion

Margaret E. I. Kipp Faculty of Information and Media Studies University of Western Ontario [email protected]

8th Information Architecture Summit, Las Vegas, Nevada. March 22-27, 2007

Background

● My exam ines: – how people organise things on the web – how this com pares to traditional library classification techniques ● Specific points of interest: – structures and the creation of structures in classification system s – relationship between personal inform ation m anagem ent and classification

What is Tagging?

● the act of associating a term with a link or article

● labelling or classifying for personal use

● act of generating a dynam ic taxonom y or folksonom y

● Related definitions: – folksonom y - user generated taxonom y of related tags – tag cloud - tag display where size equals

popularity What is Social Bookm arking?

● public sharing of links – association of tags (keywords) with links ● network of related links created by users – network of related tags created by users ● site for sharing bookm arks, articles, etc. – tags and articles are joined into networks of related term s – users are encouraged to share bookm arks and tags with others

A Social Bookm arking Post

A post is a relationship between a user, an item and a set of tags.

The Sites

Site Details

For anyone Specialised for academ ics ● del.icio.us ● – bookm ark anything citeulike (e.g. web page, – focuses on journals PDF, video clip, and academ ic audio, etc) books – now owned by – http://citeulike.org/ Yahoo! ● – http://del.icio.us/ – accepts non academ ic links

– http://connotea.org Previous Studies

● Study 1: Del.icio.us ● Study 2: Citeulike

● tag usage ● citeulike tag usage on highly tagged com pared to sites author keywords and subject ● exam ination of convergence of tag headings from usage journal databases ● exam ine types of ● co-occurrence analysis for co- tags and m ore used tags traditional index term s Non Subject Tags

● Study 1: Del.icio.us

● Affective Tags – cool: 906 occurrences ● Tim e and Task Tags – toread: 939 occurrences ● 3049 unique tags identified as tim e and task (16%)

Del.icio.us Tag Cloud

Non Subject Tags 2

● Study 2: Citeulike – average of 3.5 per article, not all from tags ● Categories: – Tim e and Task Managem ent (@toread) – Geographic Tags (berkeley) – Specifics (pubm ed-m ining) – Generalities (inform ation science) – Em ergent Vocabulary (folksonom y) – Other (system defined placeholder no-tag)

Motivations

● Previous studies showed use of non subject tags (in del.icio.us study they were at least 16% of the sam ple!)

● What is the purpose of these tags?

● Are they only individual or is there any collective useful inform ation here?

Research Questions

● What patterns of user tagging activity em erge on exam ination of affective or tim e and task related tags?

● How do users use tim e and task related tags or affective tags to indicate the value they see in a docum ent?

● What im plications do the use of affective or tim e and task related tags have for the organisation of inform ation?

Methodology

● Data source – Del.icio.us, Citeulike, Connotea ● Collection tim es – October 20-31st, 2006 ● Collection m ethod: – python scripts ● Data collected – all item s tagged with a specific list of 83 affective and tim e and task related tags

Exam ples of Tags Collected

● Affective Tags ● Tim e and Task – cool Tags – boring – @toread – exciting – todo – im portant – tobuy – funny – toblog – strange – tovisit – favorite/favourite – @pending – todescribe

Data Collected

● all posts tagged with one of 83 specific tags – 48 tags were tim e and task related, 30 were affective tags and the rem aining 5 consisted of prepositions and conjunctions – 78 tags were in English – 5 tags were in French (lire, alire, @lire, acheter, am usant)

● non English tags do not yet appear frequently in the popular tag clouds of these sites

Data Collected

● total of 203 352 posts collected – 1831 posts were collected from Citeulike – 2891 from Connotea – 198630 from Del.icio.us ● without zeros, total of 107 584 posts – 1805 for Citeulike – 2462 for Connotea – 107 584 for Del.icio.us

General Results

● som e tim e and task or affective tags are very popular – cool, fun, funny, toread appeared in m ain del.icio.us tag cloud ● affective term s appear m ore frequent on Citeulike and Connotea than Del.icio.us when norm alised for the size of the respective populations

General Results 2

● ToRead and fun are popular tags on all three sites

● prepositions and conjunctions were surprisingly popular (and appears in the citeulike tag cloud and of in the connotea tag cloud) – exam ination of the tag lists from which these tags occur suggests that the use of these prepositions is by people who are using phrases to tag an item instead of individual

words or m eaningful com pound words Popular Non-Subject Tags Non Subject Tags - Citeulike 225 200 175 150 125 100 75 50 25 0 l f t t r y k s e y s e s g n d d o d n g d d n o d i i i o o u t r n t o k s a n g a p a d n u n o n f d m i i a i p o o f o g c e e e e w n n a t t o t t c d i u k r r t a r n a w o u s t R i a c c f r n h o e h s t o e p x e r u e p r s T f e p h e i t l a m c i n h

i Popular Non-Subject Tags Non Subject Tags - Connotea 325 300 275 250 225 200 175 150 125 100 75 50 25 0 l t f t r y y s e s k e s g d d o d d d n g d n n o i o i i t n o u t r o s k n g a a p a n n d u n o f d m i o i a p i f o g o e c e e e w n n a t o t t t c d k i u r r t a r n a o w u s R t i a c c f r n h o e h s o t p e e x u r e p r s T f p e h e i t l a m c i n h

i Popular Non-Subject Tags Non Subject Tags - Del.icio.us 10000 9000 8000 7000 6000 5000 4000 3000 2000 1000 0 l f t t r y y e s e s s k g n n o g d d n d d o d d i o i i o u t t n r o k s g n p a a a n u o n n d f d m i i o a i p f o g o c e e e e w n t n a t t t o c d i u k r r t r a n a o w s u t R i a c c f r n o h e h s t o p e x e r u e p r s T f e p h e i t l m a c i n h

i Com paring Tag Frequency

Normalised Tags - Non Zero 13 12

y Citeulike

c 11

n Connotea e 10 u Del.icio.us q

e 9 r F

8 g

a 7 T 6 d e

s 5 i l

a 4 m

r 3 o

N 2 1 0 l t y s e e s o n g d o d d d o i n t s n a g p a a u n d m i o i p a f e e e e w n n t o t t c d u r r t r n a o u s t R i a f r n o e h s o t p e u r p r s T p e t m a i n h i

Popular non subject related tags com pared by site. Tag counts have been norm alised to allow com parison. Tim e and Task Tags

● m any tim e and task related tags are variations on toread – @toread – @read – readlater – unread ● fewer variations for toread are found on citeulike and connotea

● citeulike offers a way to m ark interest in reading an article Tim e and Task 2

● is the toread tag useful to other users?

● Am azon's recom m endation system relies on purchase data as an indicator of interest

● could a toread tag have a sim ilar function?

● could this function like a colleague's e- m ail pointing to an article or book?

Personal Inform ation Managem ent

● tim e and task related tags suggest a connection to research into personal inform ation m anagem ent

● research into how people organise their docum ents shows the use of contextual project inform ation for classification

● project and task related tags were found in previous studies – e.g. tags which appeared to be course codes (lis501)

Affective Tags

● affective tags represent an em otional reaction to an item – cool – fun – strange ● tags such as cool or fun do not appear to add anything to the subject classification of an item

● seem to be poor candidates for search

term s for inform atio n retrieval Affective Tags 2

Normalised Frequency of Affective Tags 14.00 12.00 Citeulike 10.00 Connotea 8.00 Del.icio.us 6.00 4.00 2.00 0.00 l t y g n e g d o i n ss n n u n g o i i p f a e t n n t i c u r n st u t a i c f o r e s x p r p st e p e t m a i n h i fun: Citeulike - 225, Connotea - 253, Del.icio.us - 9659 Non Subject Tags with Subject Tags

● non subject tags were frequently used with subject related tags

● academ ic articles on citeulike and connotea were tagged with term s such as fun and cool

● what do affective tags add to a tag list?

Non Subject Tags: Citeulike

● Title: Sym m etry and Self-Organization in Com plex System s

● URL: http://arxiv.org/pdf/cond- m at/0609274

● taglist: autom ata, fun, graphs, m athem atics, networks, statistical- m echanics, sym m etry

Non Subject Tags: Connotea

● Title: Foundations for engineering biology

● URL: http://www.nature.com /nature/journal/v43 8/n7067/full/nature04342.htm l

● Journal: Nature 438(7067): 449-453 (doi:10.1038/nature04342)

● taglist: com plex system s, network, system s biology, synthetic biology, com ics, fun

Non Subject Tags: Del.icio.us

● Del.icio.us:

● Title: 36 Hum orous Proof Methods

● URL: http://www.them athlab.com /geom etry/fun nyproofs.htm

● taglist: fun, hum or, m ath, proof

toread with Subject Tags y

c toread With Other Tags n

e 0.8 u q

e 0.7 r citeulike F 0.6 connotea e

g del.icio.us

a 0.5 t n

e 0.4 c r

e 0.3 P

d 0.2 e s i l 0.1 a m

r 0 o y s y s n g h r t N c c l g t o n i i i i t s o t a r s i l a a e y e o t i m h h e m u b e p n e i p h h g c t m n a o e c m biology: citeulike - 2, connotea - 23, del.icio.us - 2 9; m athem atics: citeulike - 0, connotea - 1, del.icio.us - 14 fun with Subject Tags

y Fun With Other Tags c

n 4.5 e

u 4 q citeulike e r 3.5 connotea F

t 3 del.icio.us n e

c 2.5 r

e 2 P

d 1.5 e s i

l 1 a 0.5 m r

o 0 N y h n r t cs cs l t o i i i t t a e s a a e y t m h h m u o p e p e h g t m a co m physics: citeulike - 82, connotea - 3, del.icio.us - 82; m athem atics: citeulike - 65, connotea - 3, del.icio.us - 20 Discussion

● non subject tags are intrinsically tim e- sensitive

● express response from user not subject of docum ent

● suggest active engagem ent with the text

● show that user links perceived subject m atter to:

● specific task

● specific set of interests

● specific em otional reactions Discussion 2

● use of tim e and task or affective tags shows that tagging expresses a dynam ic relationship between users and docum ents, suggesting possible new ways of m odelling inform ation access

● research into personal inform ation m anagem ent system s show users classify by task and project as well was by subject

Final Thoughts

● What is the effect of personal and subjective term s such as cool, fun and toread in a social bookm arking system ?

● What happens when these term s are aggregated?

● Am azon and Google use personal inform ation to generate popularity or relevance indicators, do non subject tags offer any sim ilar advantages?

Margaret E. I. Kipp Faculty of Inform ation and Media Studies University of Western Ontario m argaret.kipp@gm ail.com http://publish.uwo.ca/~ m kipp/

Thank you!

Questions?

8th Information Architecture Summit, Las Vegas, Nevada. March 22-27, 2007