Semantic Heterogeneity in Global Information Systems

The Role of Metadata Context and Ontologies

Vipul Kashyap and Amit Sheth

LSDIS Dept of Computer Science Univ of Georgia Athens GA

Dept of Computer Science Rutgers University New Brunswick NJ

June

Abstract

Semantic heterogeneity has b een identied as one of the most imp ortant and toughest prob

lems when dealing with interop erabil i ty and co op eration among multiple databases It was

earlier studied in the context of exchanging sharing and integrating data esp ecially during the

schemaview analysis phase of schema or view integration or when writing a view or query

using a multidatabase language With the adventofglobalinterconnectivi tywenowneedto

deal with more heterogeneous information resources consisting of a variety of digital data and

the scale of the problem has changed from a few databases to millions of information resources

thus making it more imp ortant than ever to address this problem It is also recognized that the

problem has only b ecome harder and that simplistic solutions involving only representational

or structural comp onents of data will not work b eyond a very restricted set of cases

In this chapter we explore approaches to tackle the semantic heterogeneity problem in the

context of Global Information Systems GIS which are systems geared to handle information

requests on the Global Information Infrastructure GI I These approaches are based on the

capture and representation of metadata contexts and ontologies In order to handle infor

mation overloaditwould b e advantageous to abstract out the representational details of the

underlying data and capture the information contentby using domain specic metadataThe

next imp ortant step is that of understanding the context of the query using metadata to con

struct the context and identifying the relevant data in that context Another critical issue that

arises here is that of dierent vocabularies used to characterize similar information We present

an approach to deal with this problem at the metadatacontext level by using terms from do

main specic ontologies to construct metadatacontext We deal with semantic heterogeneity

at this level and prop ose an approach using terminological relationships to achievesemantic

interop erabili ty

Intro duction

Many organizations face the challenge of interop erating among multiple indep endently develop ed

database systems to p erform critical functions Three of the b est known approaches to deal with

multiple databases are tightlycoupled federation lo oselycoupled federation and interdep endent

data management SLShe A critical task in creating a tightlycoupled federation is that of

schema integration eg DH A critical task in accessing data in a lo oselycoupled federation

LA HMistodeneaviewover multiple databases or to dene a query using a multidatabase

language The problem of semantic heterogeneitywhich is dened in SKas identication of

semantically related objects in dierent databases and the resolution of schematic dierences among

them is a critical issue in any of the ab ove three tasks

However with global interconnectivitywenow need to deal with more heterogeneous informa

tion resources consisting of a variety of digital data Huge amounts of digital data in a variety

of structured eg relational databases semistructured eg email messages and unstructured

App ears as a chapter in Cooperative Information Systems Current Trends and DirectionsMPapazoglou

and G Schlageter editors

eg image data formats have b een collected and stored in thousands of autonomous rep ositories

and CDROMs Aordable multimedia systems allow creation of multimedia data and supp ort ac

cess and presentation of such data These digital rep ositories are increasingly b eing made available

on the fast evolving GI I of which the World Wide Web BL is an oftcited and p opular example

A GIS now has to deal with millions of information resources as opp osed to a few databases in

amultidatabase federation and simplistic solutions involving only representational or structural

comp onents of data will not work b eyond a very restricted set of cases

In this chapter we explore approaches that use metadata context and ontologies to handle the

semantic heterogeneity problem in a GIS Two basic comp onents of these approaches are Figure

Use of metadata to capture the information content of the data in the underlying rep ositories

Intensional descriptions constructed from metadata and termed as metadata contexts m

contexts are used to abstract from the structure and organization of the individual rep ositories

Terms concepts roles in domain sp ecic ontologies are used to characterize contextual de

scriptions and are called conceptual contexts ccontextsSemantic interop erabilityisachieved

by using terminological relationships b etween terms across ontologies

ONTOLOGICAL-TERMS VOCABULARY (Domain-specific, application-specific)

used-by used-by METADATA/CONTEXT CONTENT (intensional descriptions)

abstracted-into abstracted-into

REPOSITORIES

STRUCTURE (Autonomous, Multiple formats)

Figure The basic comp onents of our approaches

The key objective of such approaches should b e to reduce the problem of knowing the contents and

the structure of each of the huge numb er of information rep ositories to the signicantly smaller problem

of knowing the contents of the domain sp ecic ontologies which a user familiar with the domain is

likely to knowor easily understand In this chapter we demonstrate the need for techniques which

go b eyond the structural and representational comp onents of data and fo cus on the application of

those techniques to structured databases

Dierenttyp es of metadata may b e stored in the system eg indices schema information

The Rufus SLS and the InfoHarness SSKS systems use automatically generated metadata

to access and retrieve heterogeneous information indep endentoftyp e representation and lo cation

In Section we discuss the dierent kinds of metadata and present an informal classication We

identify and prop ose domain specic metadata as the key for solving the semantic heterogeneity

problem KSS

Section discusses the construction of ccontexts from domain sp ecic ontologies and their rep

resentation in a formalism that can b e easily mapp ed KS to a description logic DL expression

Issues of language and ontology involved in the ab ove are also discussed These ccontexts are used

to represent extraknowledge ab out the information content of the database whichmay not b e repre

sented in the schema of the database A user query can also b e represented as a ccontext Schema

correspondences KS that capture the asso ciations b etween ccontexts and the underlying data

are also discussed

The key to interop erabilityis vocabulary sharing among the intensional mcontext and ccontext

descriptions asso ciated with the various databases Dierent concepts may b e used to design con

textual descriptions for dierent databases We assume the existence of application and domain

specic ontologies describing the information contentofthevarious databases from which contex

tual expressions may b e constructed In fact ontologies are viewed in our approach as a sp ecial case

of domain specic metadata In Section we present an approach for semantic interop erability us

ing terminological relationships across ontologies We discuss the OBSERVER prototyp e MKSI

which demonstrates the use of synonym relationships to achieve semantic interop erability Exten

sions of the ab oveusinghyponym and hypernym relationships MKIS are also discussed Section

presents future and ongoing work and our conclusions

What is Metadata

Figure illustrates the twocomponents of our approach for addressing the information overload

problem in the GI I Metadata is the pivotal idea on which b oth the comp onents dep end The func

tion of the metadata descriptions is to b e able to abstract out and capture the essential information

in the underlying data independent of representational details This represents the rst step in re

duction of the information overload as metadata descriptions are in general an order of magnitude

less in size than the underlying data In this section we discuss in detail our notion of metadata

the various typ es of metadata and the information they capture

Metadata in its most general sense is dened as data or information ab out data For structured

databases the most common example of metadata is the schema of the database However with the

proliferation of various typ es of multimedia data on the GI I we shall refer to an expanded notion

of metadata of which the schema of structured databases is a small part We use metadata to

store derived prop erties of media useful in information access or retrieval They may describ e or b e

a summary of the information content of the data describ ed in an intensional manner They may

also b e used to represent prop erties of or relationships b etween individual ob jects of heterogeneous

typ es and media Wenow discuss a classication of the dierenttyp es of metadata and characterize

the amount of information content they capture Wealsoidentify the typ es of metadata whichwill

playa key role in enabling semantic interoperability

A Classication of Metadata

Wenow present a classication of the various typ es of metadata used byvarious researchers to

capture the information content represented in the various typ es of digital data Table

Content Indep endent Metadata This typ e of metadata captures information that do es not de

p end on the content of the do cument with which it is asso ciated Examples of this typ e of

metadata are location modicationdate of a do cument and typeofsensor used to record a

photographic image There is no information content captured by these metadata but these

might still b e useful for retrieval of do cuments from their actual physical lo cations and for

checking whether the information is upto date or not

Content Dep endent Metadata This typ e of metadata dep ends on the content of the do cument

it is asso ciated with Examples of content dep endent metadata are size of a do cument max

colors numberofrows numberofcolumns of an image Wenowpresent a categorization of

content dep endent metadata

Direct Contentbased Metadata This typ e of metadata is based directly on the contents

of a do cument A p opular example of this is fulltext indices based on the text of the

do cuments Invertedtree and document vectors are examples of this typ e of metadata

Contentdescriptive Metadata This typ e of metadata describ es the contents of a do cu

ment without direct utilization of the contents of the do cument An example of this typ e

of metadata is textual annotations describing the contents of an image This typ e of

metadata comes in twoavors

Domain Indep endent Metadata These metadata capture information present in the

do cument indep endent of the application or sub ject domain of the information Ex

amples of these are the CC parse trees and HTMLSGML document type de

nitions

Domain Sp ecic Metadata Metadata of this typ e is describ ed in a manner sp ecic

to the application or sub ject domain of information Issues of vo cabulary b ecome

very imp ortant in this case as the terms havetobechosen in a domain sp ecic

manner Examples of such metadata are relief landcover from the GIS domain

and area population from the Census domain In the case of structured data the

database schema is an example of such metadata Another interesting example is

domain specic ontologies terms from whichmaybeusedasvo cabulary to construct

metadata sp ecic to that domain

Metadata Media Typ e Metadata Typ e

QFeatures JH Image Video Domain Sp ecic

RFeatures JH Image Video Domain Indep endent

RFeatures JH Image Video Content Indep endent

Impression Vector KKH Image Content Descriptive

NDVI Spatial Registration AS Image Domain Sp ecic

Sp eech feature index GSW Audio Direct Contentbased

Topic change indices CHK Audio Direct Contentbased

Do cumentVectors DDF Text Direct Contentbased

Inverted Indices KM Text Direct Contentbased

Content Classication Metadata BR MultiMedia Domain Sp ecic

Do cument Comp osition Metadata BR MultiMedia Domain Indep endent

Metadata Templates OM Media Indep endent Domain Sp ecic

LandCover Relief SK Media Indep endent Domain Sp ecic

ParentChild Relationships SSKS Text Domain Indep endent

Contexts SSR KSa Structured Databases Domain Sp ecic

Concepts from Cyc CHS Structured Databases Domain Sp ecic

Users Data Attributes SLS Text Structured Databases Domain Sp ecic

Domain Sp ecic Ontologies MKSI MediaIndep endent Domain Sp ecic

Table Metadata for Digital Media

Metadata A means for capturing information content

In this section we discuss the information content captured bythevarious typ es of metadata enu

merated in the previous section We shall also identify the level Figure at which this metadata

may b e used

Content Indep endent Information This typ e of information is captured byContent Indep en

dent metadata and helps in the encapsulation of information into units of interest and maybe

represented as ob jects in a data mo del

Capturing Representational Information This typ e of information is typically captured by

Content Dep endent Metadata describ ed in the previous section This along with Domain

Indep endent Metadata which primarily captures structural organization of the data enables

interop erabilityvianavigational and browsing approaches which dep end on representational

details of the data

Capturing Information Content Information Contentistypically captured to various degrees

byvarious typ es of Content Dep endent Metadata Direct Contentbased Metadata lies in a

grey area in the sense that it is not entirely divorced from the representational details How

ever the metadata which helps abstract out representational details and capture information

meaningful to a particular application or sub ject domain is Domain Sp ecic Metadata

Vo cabulary for Information Content Characterization Domain Sp ecic Metadata can b e con

structed from terms in a domain sp ecic ontology or concept libraries describing information

in an application or sub ject domain Thus weviewOntologies as Metadata which themselves

can b e viewed as a vo cabulary of terms for construction of more domain sp ecic metadata

descriptions Semantic interop erabilityatthevo cabulary level is achieved with the help of

terminological relationships

The ab ove discussion suggests that domain specic metadata capture information whichismore

meaningful wrt a sp ecic application or a domain The information captured by the other typ es

of metadata primarily reect the format and organization of the underlying data This leads us to

prop ose domain sp ecic metadata as the most appropriate for dealing with issues related to semantic

heterogeneity

Constructing Intensional Descriptions from Domain Sp ecic Metadata

Domain sp ecic metadata can b e used to construct intensional descriptions which capture the in

formation content of the underlying data We categorize these intensional descriptions as follows

Metadata Contexts mcontexts These descriptions primarily serve to abstract the represen

tational details in the underlying data and may b e viewed as boolean combinations of the

individual metadatum These contexts are typically p opulated b efore hand by pro cessing the

underlying data They may also b e computed at runtime byusingparameterizedroutinesEx

amples of this typ e of metadata and howtheymaybeusedtointerop erate across multimedia

data are illustrated in SK

Conceptual Contexts ccontexts These descriptions primarily servetocapturedomain know l

edge and help imp ose a conceptual semantic view on the underlying data Ccontexts are

constructed from terms concepts roles in domain sp ecic ontologies The terms used in

construction of the ccontexts mightbeinterrelated to each other via relationships viz termi

nological domainrange constraints on roles

In the rest of the chapter we fo cus on the structured data and the use of ccontexts constructed

from domain sp ecic ontologies to capture the information content The relationships b etween

terms in the ontologies enable the representation of extraknowledge not represented in the database

schema We shall also discuss the cases where ccontexts may b e constructed from dierent domain

sp ecic ontologies

Constructing ccontexts from Ontological Terms

In Figure wehaveidentied metadata as the pivotal idea on which our approaches to address the

information overload problem in the GI I are based In the previous section we discussed the various

typ es of metadata and identied domain sp ecic metadata as the most appropriate for handling

semantic heterogeneity One approach to construct metadata which capture meaningful information

wrt an application domain is to use terms from domain sp ecic ontologies as the vo cabulary to

characterize the information Wehave identied such metadata descriptions as ccontexts in the

previous section and in this section we present a discussion of issues related to their representation

and use

We discuss the inadequacies of purely structural and mapping based metho ds in representing

ob ject relationships and discuss the advantages of representing ccontexts We shall discuss a partial

representation of ccontexts and equivalent description logic expressions We shall also discuss

op erations for automatic ways of comparing and manipulating ccontexts and illustrate with the

help of examples how they mayb e used to achieveinterop eration across information sources A

brief discussion of issues relating to the language for representing ccontexts and the ontologies from

which the ccontexts may b e constructed is also given We shall refer to ccontexts as contexts

unless otherwise sp ecied in the rest of the chapter

Rationale for Context representation

In characterizing the similaritybetween ob jects based on the semantics asso ciated with them we

have to consider the real world semantics RWS of an ob ject It is not p ossible to completely dene

what an ob ject denotes or means in the mo del world SG We prop ose the context of an ob ject as

the primary vehicle to capture the RWS of the ob ject We argue for the need for representing context

by showing the inadequacy of purely structural representations We also discuss the computational

b enets of representing context

Inadequacy of purely Structural Representations

It has b een suggested by Sheth and GalaKashyap SGSKandFankhauser et al FKN

that the ability to represent the structure of an ob ject do es not help capture the real world semantics

of the ob ject It is not p ossible to provide a structural and hence a mathematical denition of the

complex notion of real world semantics In LNE a onetoone mapping is assumed b etween

the attribute denition and the attributes real world semantics They dene an attribute in terms

of xed descriptors suchasUniqueness LowerUpper Bound Domain Scale etc which are used

to generate mappings b etween two attributes They are also used to determine the equivalence of

attributes However what they establish is the structural equivalence of these attributes whichis

necessary but not sucient to determine the semantic equivalence of the attributes

Consider two attributes personname and departmentnameWemay b e able to dene a map

ping b etween the value domains of these two attributes but weknowthattheyarenotsemantically

equivalent In order to b e able to capture this lack of equivalence we prop ose the mappings b etween

the domains of the attributes b e made wrt acontext We dene two ob jects to b e semantically equiv

alentifit is p ossible to dene mappings wrt all known and coherent contexts and the denition contexts

of the objects should b e coherent wrt each other Denition contexts and the notion of coherence are

discussed later in this section Since the denition contexts of personname and departmentname

are not coherent one identies an animate and the other identies an inanimate ob ject they are

not dened to b e equivalent

Computational b enets of representing context

Shoham Sho has discussed the computational b enets that might accrue in mo deling and rep

resenting context in AI and KnowledgeBased systems We b elieve that some of those reasons are

very relevant in the presence of information overload in the GI I and suggest the identication and

representation of context

Economy of representation In a manner akin to database views contexts can act as a focusing

mechanism when accessing the comp onent databases on the GI I They can b e a semantic

summary of the information in a database or group of databases and mayb e able to capture

semantic information not expressed in the database schemas Thus unnecessary details can

b e abstracted from the user

Economy of reasoning Instead of reasoning with the information present in the database as a

whole reasoning can b e p erformed with the context asso ciated with a database or a group

of databases This approach has b een used in KSa for information resource discovery and

query pro cessing in Multidatabases

Managing Inconsistent Information In the GI I where databases are designed and develop ed

indep endently it is not uncommon to have information in one database inconsistent with

information in another As long as information is consistent within the context of the query

of the user inconsistency in information from dierent databases may b e allowed This has

b een discussed with the help of an example in KS

Flexible semantics An imp ortant consequence of asso ciating abstractionsmappings with the

context is that the same two ob jects can b e related to each other dierently in two dierent

contexts This is b ecause two ob jects mightbesemantically closer to each other in one context

as compared to the other

A partial Context representation

There have b een attempts to represent the similaritybetween ob jects in dierent databases In

the previous section weshowed with the help of an example how a xed set of descriptors used in

LNE do not guarantee semantic similarityThus any representation of context whichcanbe

describ ed by a xed set of descriptors is not appropriate

The descriptors called metaattributes or contextual co ordinates are not xed but are dynam

ically chosen to mo del the characteristics of the application domain in question It is not p ossible

apriori to determine all p ossible contextual co ordinates whichwould completely characterize the

semantics of the application domain This leads to a partial representation of context as a collection

of contextual co ordinates

Context
C V C V C V 

    k k

Table shows how our context descriptions can b e mapp ed to expressions in CLASSIC BBMR a



DL system Using CLASSIC it is p ossible to dene primitive classes and in addition sp ecify classes

using intensional descriptions phrased in terms of necessary and sucient prop erties that must b e

satised by their instances The intensional descriptions may b e used to express the collection of

constraints that make up a context Also eachC roughly corresp onds to a role and eachV roughly

i i

corresp onds to llers for the role the ob ject must have We shall also explain the meaning of the

symbols C and V by using examples and byenumerating the corresp onding CLASSIC expressions

i i

C i k is a contextual co ordinate denoting an asp ect of a context

i

C may mo del some characteristic of the sub ject domain and may b e obtained from a domain

i

sp ecic ontology discussed later in this section

C may mo del an implicit assumption in the design of a database

i

C mayormay not b e asso ciated with an attribute A of an ob ject O in the database

i j

Contextual co ordinates and Values CLASSIC descriptions

C V C V  AND OALL C V ALL C V

  k k   k k

C O
C V  AND OALL C AND O ALL C V

i i j j i j j j

C XC X C forSAMEAS C C

i j i i j

C X
C V  C forFILLS C ALL C V

i j j i i j j

Table Contextual co ordinate value pairs and the corresp onding CLASSIC expressions

The value V of a contextual co ordinate C can b e represented in the following manner

i i

V can b e a variable

i

It can b e unied in the sense of Prolog with another variable a set of symb ols an

ob ject or typ e dened in the database or another variable

It can b e unied with another variable asso ciated with a context

It can b e used as a place holder to elicit answers from the databases and imp ose con

straints on them

Wehave prop osed a minor addition rolesetfor classicexpression to CLASSIC expressions MKSIto

enable retrieval of ob ject prop erties

Example

Supp ose we are interested in p eople who are authors and who hold a p ost We can represent

the query context C discussed later in this section as follows

q

C
author X designee X

q

The same thing can b e expressed in a Description Logic DL as follows

C author for SAMEAS author designee

q

The terms author and designee may b e roles chosen from a domain sp ecic ontology

V can b e a set

i

The set maybeanenumeration of symb ols from a domain sp ecic ontology

The set may b e dened as the extension of an ob ject or as elements from the domain of

atyp e dened in the database

The set may b e dened by p osing constraints on preexisting sets

Example

Supp ose wewanttorepresent the assumptions implicit in the design of the ob ject EM

PLOYEE in a database We can represent this as the denition context of EMPLOYEE

C EMPLOYEE as follows

def

C EMPLOYEE
employer Deptyp es frestyp esgarticle PUBLICATION

def

The same thing can b e expressed in a DL as follows

C EMPLOYEE AND EMPLOYEE ALL article PUBLICATION

def

ALL employer Deptyp es frestyp esg

Deptyp es is a typ e dened in the database The symbols restypes employer and article

maybechosen from a domain sp ecic ontologyThesymbols employer and article maybe

related to attributes asso ciated with the underlying database ob jects The symbol restypes

acts as a role ller and maybemappedtoadatavalue in the database The denition context

expresses an asso ciation b etween the ob jects EMPLOYEE and PUBLICATION whichmay

not b e captured in the database schema

V can b e a variable asso ciated with a context

i

This can b e used to express constraints which the result of a query should ob ey and is

called the constraint context

The constraints would apply to the set typ e or ob ject the variable X would unify with

Example

Supp ose wewant all articles whose titles contain the substring ab ortion in them This can

b e expressed in the following query context

C
article X
title fyjsubstringy ab ortiong

q

where denotes association of a context
title fyjsubstringy ab ortiongwitha

variable X and ensures that the answer satises the constraints expressed in the context

The same thing can b e expressed in a DL as follows

C article for ALL title fyjsubstringy ab ortiong

q

V can b e a set typ e or an ob ject asso ciated with a context This is called the asso ciation

i

context and may b e used to express semantic dep endencies b etween ob jects whichmaynotbe

mo deled in the database schema

Example

Supp ose wewant to represent information relating publications to employees in a database

Let PUBLICATION and EMPLOYEE b e ob jects in a database The denition context of

HASPUBLICATION can b e dened as

C HASPUBLICATION
article PUBLICATION

def

author EMPLOYEE
aliation fresearchg

where denotes asso ciation of a context with an ob ject EMPLOYEE and a context

aliation fresearchg

Association of a context with an ob ject is similar to dening a view on the ob ject extensions

such that only those instances satisfying the constraints dened in the context are exp orted

to the GIS The symb ols used as contextual co ordinates eg article author and aliation

are obtained from a domain sp ecic ontology and may b e mapp ed to attributes of database

ob jects The relationships b etween the database ob jects EMPLOYEE PUBLICATION and

HASPUBLICATION captured in the contextual description are not mo deled in the database

schema The same thing can b e expressed in a DL as follows

C HASPUBLICATION AND HASPUBLICATION

def

ALL article PUBLICATION

ALL author AND EMPLOYEE

ALL aliation ONEOF fresearchg

Reasoning ab out and manipulation of contexts

Wehave prop osed a partial representation of context in the previous section This can b e used

to abstract out the information content of the underlying data and help reduce the information

overload in the GI I The next step is to use these representations meaningfully to enable a GIS to

fo cus on relevant information and to correlate information from the various information sources on

the GI I In order to achieve this the following need to b e precisely dened KS

Sp ecicity The most common relationship b etween contexts is the sp ecicity relationship Given

two contexts C and C C C i C is at least as sp ecic as C This is useful when ob jects

     

dened in a particular context have to transcend McC to a more sp ecic or general context

and is discussed in detail with examples in KS

Organization in a Lattice Structure It is p ossible that twocontexts may not b e comparable to

each other ie it may not b e p ossible to decide whether one is more sp ecic than the other

Thus the sp ecicity relationship gives us a partial order The following useful op erations on

the context lattice can b e dened

overlapCntxt Cntxt This is the common set of contextual attributes present in the

 

contextual descriptions

coherentCntxt Cntxt This op erator determines whether the constraints determined

 

by the values of the contextual co ordinates are consistent

Example

Let Cntxt
salary fxj x g



Cntxt
salary fxj x  g



Thus coherentCntxt Cntxt FALSE

 

greatest lower b ound glb of two contexts The contexts can b e organized in a sp ecial

kind of lattice structure called a meet semilattice in whichevery pair of contexts has

a greatest lower b ound Intuitively the glb computes the conjunction of constraints ex

pressed in the contextual descriptions

Inferences using Contextual Descriptions

Wenow illustrate how reasoning with contextual descriptions can help enable semantic interop er

ability across dierent databases on the GI I The interop erabilityisachieved wrt the query which

is represented as a context and known as the query context C The denition contexts of the

Q

various ob jects in the underlying databases enable the partial capture and representation of the

information content in the databases The query context is compared with the denition contexts

and this can b e easily implemented as a combination of the glb and overlap op erations discussed

ab ove

A critical assumption made in the examples illustrated b elow is that query and denition contexts

are constructed from a common ontologyThisisavery unscalable assumption in the context of a

GIS One way of enhancing the scalability is to supp ort the use of preexisting and indep endently

develop ed often adho c domain ontologies This requires mechanisms for comparing terms across

ontologies at runtime which is the sub ject of discussion of the next section Issues of language to

represent the contextual descriptions and ontologies are discussed later in this section

Consider the comparison of the query context C and the denition context C PUBLICATION

Q def illustrated in Figure

C Cdef(PUBLICATION) Q <(author,X)(designee,X) <(researchArea, Department)> (employer, Department) (article,Yo<(title,{x | contains(x,"abortion")})>) (researchArea,{socialSciences,politicalSciences})>

compare(Cdef(PUBLICATION), CQ)

<(researchArea, {socialSciences, politicalSciences})>

Figure Comparison of contextual descriptions Identifying the relevant publications

The instances of the PUBLICATION ob ject identied as b elonging to the research areas so

cialSciences and politicalSciences are determined to b e relevant to the user query This is an example

of using contextual expressions for determining information relevanttoa query

In the next example we illustrate how constraints in a query can b e applied to information in

a database to determine the relevant answers Consider the query context C and the denition

Q

context C HASPUBLICATION illustrated in Figure def

C Cdef(HAS-PUBLICATION) Q <(author,FACULTYo<(affiliation,researchTypes)>) <(author,X)(designee,X) (article,PUBLICATION)> (employer,Department) (article,Yo<(title,{x | contains(x,"abortion")})>) (researchArea,{socialSciences,politics})>

compare(Cdef(HAS-PUBLICATION), CQ) <(author,FACULTYo<(affiliation,researchTypes)>)

(article,PUBLICATIONo<(title,{x | contains(x, "abortion")})>)>

Figure Comparison of contextual descriptions Incorp orating a constraint from the query

The constraint in the query requiring the article titles to contain the word ab ortion is in

corp orated in the contextual descriptions describing the information content of the database and

propagated to the ob ject PUBLICATION The mo died contextual description thus characterizes

only those instances of the ob ject PUBLICATION whichcontain the word ab ortion in their titles

Another interesting use of contextual descriptions is to rule out the p ossibility of a database

having information relevant to a query Supp ose we are interested in all authors having a salary 

Supp ose all the faculty memb ers in the university database are represented as having a

salary Consider the following contextual descriptions

C FACULTY
salary fxj x g

def

C
author X salary fxj x  g

Q

compareC FACULTYC inconsistentx x 

def Q

The university database is not relevant for the query Q

Mapping Contextual descriptions to the Database Schema

As discussed earlier the contextual descriptions serve to abstract out the underlying representa

tional details and capture the information content However once the relevant highlevel contextual

descriptions have b een identied there is a need to retrieve the relevant data and display it to the

user In KS we prop ose a uniform formalism used to map contextual descriptions to underlying

data Work on mapping intensional descriptions to SQL queries is rep orted in BB Collet et al

CHShave used articulation axioms to relate ob ject classes in databases to concepts in the Cyc

ontology Our approach is similar to the ab ovebutwehave also dened an algebra in KSto

keep track of the changes in the mappings when the asso ciated contextual descriptions change

Each information system exp orts a global ob ject O corresp onding to the ob jects O it man

G

ages to the GIS The ob jects O are obtained by applying the constraints in the denition context

G

C O to the ob ject O The user sees only the exp orted ob jects The contextual co ordinates C of

def i

the C O act as the attributes of O The exp orted ob jects O are asso ciated with the ob jects

def G G

and typ es dened in the database This asso ciation might b e implemented in dierentways by

various comp onent systems We use schema corresp ondences dened as follows to express these

asso ciations Figure

schCorO O
O fC j C C OgOattrOM

G G i i def

O is the exp orted GIS ob ject of an ob ject O or typ e T dened in the database

G

The attributes of the ob ject O are the contextual co ordinates of the denition context

G

C O

def

The mapping op eration map C A stores the asso ciation b etween contextual co ordinate

O i i

C and attribute A of ob ject O whenever there exists one

i i

The mapping M b etween O and O can b e evaluated using the pro jection rules enumerated

G

and illustrated in KS

FEDERATION LEVEL

GIS Object O G Attributes C1 , C2 , ..., Ck

Cdef(O) <(C1, V1) ... (Ck ,Vk )> ... mapO(Ci, Ai) DATABASE LEVEL Database O Attributes A1 , A2 , ..., Ak

Object

Figure Schema Corresp ondences Mapping contextual expressions to underlying data

Wehave discussed in KS a set of projection rules which map a contextual expression to

underlying database ob jects Wenow discuss two examples whichillustratehow extra information

may b e represented using contextual expressions

Representing relationships b etween ob jects

We illustrate a case where the denition context of the ob ject HASPUBLICATION captures its rela

tionships with another database ob ject EMPLOYEE in an intensional manner These relationships

are not stored in the database and mapping the contextual description results in extra information

b eing asso ciated with the GIS ob ject HASPUBLICATION Anaive user will ordinarily not b e

G

aware of this relationship The detailed mapping of this relationship has b een illustrated in KS

Example

Consider ob jects EMPLOYEE and PUBLICATION dened earlier and an ob ject

HASPUBLICATIONSS Id in the same database which represents a relationship b etween em

ployees and the publications they write

C HASPUBLICATION
authorEMPLOYEE
aliationfresearchg

def

HASPUBLICATION JoinSS SS HASPUBLICATION

G

SelectAliation fresearchg EMPLOYEE

This results in only those ob jects b eing exp orted to the GIS which satisfy the constraints sp ecied

in the contextual descriptions The user thus do es not havetokeep track or know the relationships

between the various ob jects in the database

Using terminological relationships in Ontology to represent extra information

In this section we illustrate an example in which terminological relationships obtained from an on

tology are used to represent extra information In the example illustrated b elow the contextual co

ordinate researchInfo is a comp osition of twocontextual co ordinates researchArea and journalTitle

and is obtained from the ontology of the domain This is then used to correlate information b etween

the ob jects PUBLICATION and JOURNAL However the contextual co ordinate researchArea has

not b een mo deled for the ob ject PUBLICATION Thus this results in extra information ab out the

relevant journals and research areas b eing asso ciated with the ob ject PUBLICATION even though

no information ab out research areas is mo deled for PUBLICATION

Example

Consider a database containing the following ob jects

PUBLICATIONId Title Journal where

C PUBLICATION

def

researchInfoJOURNAL
researchAreaDeptyp e sjournalTitleJournalTyp es

JOURNALTitle Area where C JOURNAL


def

The mapping expression is given as follows see KS for details

PUBLICATION JoinresearchAreaAreaTitleJournal PUBLCATION

G

SelectAreaDeptyp esTitleJournalTyp esJOURNAL

Only journals b elonging to the research areas corresp onding to the departments are selected

SelectArea IN Deptyp es AND JOURNAL

The join condition Title Journal ensures that only those articles which are from the

research areas corresp onding to the departments are exp orted to the GIS

JoinresearchAreaAre a AND Title Journal

This is achieved even though the attribute Area is not mo deled for PUBLICATION Thus

there is extra information in terms of asso ciation of Deptyp es with PUBLICATION through

the join condition

Issues of language and ontology in context representation

In this section we discuss the issues of a language in which the explicit context representation

discussed in Section can b e b est expressed Besides as discussed earlier we use terms from

domain sp ecic ontologies as vo cabulary to characterize domain sp ecic information We also discuss

in this section issues of ontology ie the vo cabulary used by the language to representthecontexts

Language for context representation

In Section wehave represented context as a collection of contextual co ordinates and their values

The values themselves mayhavecontexts asso ciated with them In this section weenumerate the

prop erties desired of a language to express the context representation

The language should b e declarative in nature as the context will typically b e used to express

constraints on ob jects in an intensional manner Besides the declarative nature of the language

will make it easier to p erform inferences on the context

The language should b e able to express the context as a collection of contextual co ordinates

each describing a sp ecic asp ect of information present in the database or requested bya

query

The language should have primitives for determining the subtyp e of twotyp es pattern match

ing etc in the mo del world which might b e useful in comparing and manipulating context

representations

The language should have primitives to p erform navigation in the ontology to identify the

abstractions related to the ontological ob jects in the query context or the denition contexts

of ob jects in the databases

The Ontology Problem

The choice of the contextual co ordinates C s and the values assigned to them V s is very imp or

i i

tant in constructing the contexts There should b e ontological commitments that imply agreements

ab out the ontological ob jects used b etween the users and the information system designers In

our case this corresp onds to an agreement on the terms and the values used for the contextual

co ordinates by b oth a user in formulating the query context and a database administrator for for

mulating the denition and asso ciation contexts In the example in Section wehave dened

C EMPLOYEE by making use of symb ols like employer aliation and reimbursement from the

def

ontology for contextual co ordinates and research teaching etc for the values of the contextual

co ordinates

We assume that each database has available to it an ontology corresp onding to a sp ecic domain

The denition and asso ciation contexts of the ob jects take their terms and values from this ontology

However in designing the denition contexts and the query context the issues of combining the

various ontologies arise Wenowenumerate various approaches one mighttake in building ontologies

for a GIS comprising of numerous information sources Other than the ontological commitment a

critical issue in designing ontologies is the scalability of the ontology as more information sources

enter the federation Two approaches are discussed next

The Common Ontology approach

One approach has b een to build an extensive global ontology A notable example of

global ontology is Cyc LG consisting of around ob jects In Cyc the mapping

between each individual information resource and global ontology is accomplished bya

set of articulation axioms which are used to map the entities of an information resource

to the concepts such as frames and slots in Cycs existing ontology CHS

Another approach has b een to exploit the semantics of a single problem domain eg

transp ortation planning ACHK The domain mo del is a declarative description of

the ob jects and activities p ossible in the application domain as viewed byatypical user

The user formulates queries using terms from the application domain

Reuse of Existing OntologiesClassications We exp ect that there will b e numerous

information systems participating in the GIS In this context it is unrealistic to exp ect any

one existing ontology or classication to suce We b elieve that the reuse of various existing

classications such as ISBN classication for publications b otanical classication for plants

is a very attractive alternative An example of such a classication is illustrated in Figure

These ontologies can then b e combined in dierentways and made available to the GIS

A critical issue in combining the various ontologies is determining the overlap b etween

them One p ossibility is to dene the intersection and mutual exclusion p oints b e

tween the various ontologies Wie Land Use and Land Cover Classification (USGS)

Urban Forest Land Water

Residential Industrial Evergreen Lakes Commercial Reservoirs Deciduous Mixed Streams and Canals

A classification using a generalization hierarchy

Population Area Classification (US Census Bureau)

State

County

City Rural Area

Tract Block Group Block

A classification using an aggregation hierarchy

Figure Examples of Generalization and Aggregation hierarchies for Ontology construction

Another approach has b een adopted in MS The typ es determined to b e similar bya

sharing advisor are classied into a collection called conceptAconcept hierarchy is thus

generated based on the sup erconceptsub conc ept relationship These typ es maybefrom

dierent databases and their similarity or dissimilarity is based on heuristics with user

input as required

Semantic Interop erability using Terminological Relation

ships

In Figure we illustrated how terms from domain sp ecic ontologies can b e used as vo cabularies

to characterize domain sp ecic information This is an essential comp onent of the approaches

to enable tackling the semantic heterogeneity problem on the GI I In the previous section we

discussed how terms from an ontology may b e used to construct contextual expressions and how

terminological relationships result in the representation of extra information not represented in

the database schema However there was an implicit assumption of a common ontology b ehind the

construction of the contextual expressions As discussed earlier this is a very unscalable assumption

In this section we discuss the issues involved when contextual descriptions may b e constructed from

dierent domain sp ecic ontologies We discuss how semantic interoperability maybeachieved

byinterop eration across these domain sp ecic ontologies Wenow discuss approaches to achieve

interop eration across ontologies using terminological relationships like synonyms hyponyms and

hypernyms

Using synonyms to interop erate across ontologies

In this section we prop ose an approachtointerop erate across ontologies whichhave b een expressed

using a description logic system like CLASSIC BBMR Wehave illustrated howcontextual

expressions may b e represented using description logic expressions Wenow discuss our work in the



OBSERVER MKSI system which enables interop eration across various indep endent preexisting

ontologies based on synonym relationships across terms in dierentontologies



Ontology Based System EnhancedwithRelationships for Vocabulary hEterogeneity Resolution

An architecture for interop eration

In this section we discuss an architecture for interop eration across domain sp ecic ontologies Fig ure

Data Repositories

IRM Mappings Ontology Server

Ontologies

Interontologies Terminological Query Processor User Query Relationships User Node IRM node

Component Node Component Node

Ontology Server Ontology Server

Mappings Mappings Query Processor Ontologies Query Processor Ontologies

Data Repositories Data Repositories

Figure OBSERVER An architecture to supp ort interop eration across ontologies

Query Pro cessor This comp onenttakes as input a user query expressed in DLs using terms

from a chosen user ontology It then navigates other comp onentontologies of the Global

Information System and translates terms in the user query into the comp onentontologies

preserving the semantics of the user query This may result in a partial translation of the query

at a comp onentontology It also combines the partial translations at the presentontology

with those determined at previous ontologies such that all constraints in the user query are

translated

Ontology Server The Ontology Server provides information ab out ontologies to the Query Pro

cessor It provides the denitions of the terms in the ontology and retrieves data underlying

the ontology It is resp onsible for evaluating the mappings of the contextual expressions to

the underlying data and retrieving the data which satises the constraints in the user query

Interontologies Relationships Manager IRM Synonym relationships relating the terms in

various ontologies are represented in a declarative manner in an indep endent rep ositoryThis

enables interop eration across the various ontologies

Ontologies EachOntology is a set of terms of interest in a particular information domain ex

pressed using DLs in our work They are organized as a lattice and may b e considered as

semantically rich metadata capturing the information content of the underlying data rep osi

tories The various ontologies used in OBSERVER are illustrated in the App endix

The Interontologies Relationship Manager IRM

The IRM is the critical comp onent which supp orts ontologybased interop eration It also enhances

the scalability of the query pro cessing strategy byavoiding the need for a designing a common

global ontology containing all the relevant terms in the Global Information System and b investing

time and energy for the developmentofanontology sp ecic for your needs when similar ontologies

are available Relationships b etween terms across ontologies that capture the overlapping of domains

are stored in a rep ository managed by the IRM The rep ository also includes information ab out

transformer functions which can transform values or rolellers from a domain in one ontology to

another The main assumption b ehind the IRM is that the numb er of relationships b etween terms

across ontologies is an order of magnitude smaller than the numb er of all the terms relevant to the

system

Hammer and McLeo d HMhave suggested a set of relationship descriptors to capture re

lationships b etween terms across dierent lo cally develop ed ontologies A set of terminological

relationships has b een prop osed in Mil In the OBSERVER system we discuss an approach us

ing synonym relationships We will discuss extensions to the OBSERVER system for using hyponyms

and hypernyms in the next section

Query Pro cessing in OBSERVER

Wenow discuss a query pro cessing approachthatinvolves the reuse of preexisting ontologies and

interop eration across them The query pro cessor p erforms the following imp ortantsteps

Translation of terms in the query into terms in eachcomponentontology The query pro cessor

obtains information from the IRM discussed in Section and the Ontology Server

Combining the partial translations in suchaway that the semantics of the user query is

preserved

Accessing the Ontology Server to obtain the data under the comp onentontology that satisfy

the translated query This basically amounts to the evaluation of the mappings of the con

textual expressions to the underlying database schema and has b een discussed in the previous

section

Correlation of the ob jects retrieved from the various data rep ositoriesontologies

We illustrate steps and using an example in MKSI A detailed discussion of the query

pro cessing strategy is describ ed in the same pap er Consider a contextual expression represented in

CLASSIC used for the following query

Get the titles authors do cuments and the numb er of pages of do ctoral theses dealing with meta

data and that have b een published at least once

Let us assume that there are ontologies describ ed in detail in MKSI as discussed b elow

StanfordI I This ontology is a subset of the Bibliographic Data Ontology Grudevelop ed

as a part of the ARPA Knowledge Sharing Eort httpwwwkslstanfordeduknowledge

sharing It corresp onds to the subtree under the concept reference of the Bibliographic

Data Ontology and is illustrated in App endix D

StanfordI This ontology is also a subset of the Bibliographic Data Ontology and corresp onds

to the rest of the ontology It is illustrated in App endix C

WN This ontology was built by reusing a part of the WordNet ontology Mil The

concepts in the WN ontology are a subset of terms in the hyp onym tree of the noun print

media It is illustrated in App endix B

LSDIS This ontology is a lo cal homegrown ontology which represents our view of our Labs

publications and is illustrated in App endix A

The query can b e constructed from the concepts in StanfordI I denoted as the user ontology

and represented in CLASSIC as follows

title author do cument pages forAND do ctoralthesisref FILLS keywords metadata

ATLEAST publisher

Wenowenumerate the translations of the query into the ontologies discussed ab oveandidentify

the translated and nontranslated parts

StanfordI I The query always represents a full translation into the user ontology

StanfordI There is a partial translation of the query at this ontology

Translated Part title author NULL numb erofpages for

AND do ctoralthesis ATLEAST publisher

Nontranslated Part FILLS keywords metadata

WN Terms in the query are substituted by their denitions in the ontology from which they are

chosen StanfordI I to obtain a complete translation into WN

do ctoralthesisref AND thesisref FILLS typ eofwork do ctoral

thesisref AND publicationref FILLS typ eofwork thesis

Translated Part name creator NULL pages for AND printmedia

FILLS content thesis do ctoral ATLEAST publisher

FILLS generaltopics metadata

LSDIS There is a partial translation at this ontology where the value of the roleller of the role

keywords is transformed by the transformer function b etween the roles keywords StanfordI I

and subject LSDIS

Translated Part title authors lo cationdo cument NULL for AND publications

FILLS typ e do ctoral thesis FILLS sub ject METADATA

Nontranslated Part ATLEAST publisher

Consider the partial translations of the user query at the ontologies StanfordI and LSDIS As

the intersection of the nontranslated parts of the partial translations into StanfordI and LSDIS is

empty then the intersection of b oth partial answers must satisfy all the constraints in the query

Intuitively

From StanfordI do ctoral theses ab out any sub ject whichhave b een published at least once will

b e retrieved

From LSDIS do cuments ab out metadata whichmay not have b een published will b e retrieved

The intersection of the ab ove will b e those do cuments classied as do ctoral theses ab out metadata

and have b een published at least once which is exactly the user query

After obtaining the corresp onding data for eachontology involved in the user query that data

must b e combined to give an answer to the user For eachanswer represented as a relation the

Query Pro cessor will transform the values in the format of the user ontology byinvoking the ap

propriate transformer functions obtained from the IRM After this initial step the dierent partial

answers can b e correlated since all of them are expressed in the language of the user ontologyThe

correlation plan corresp onding to the translations illustrated ab oveis

User Query Ob jects Ob jectsself title author do cument pages for AND do ctoralthesisref FILLS

keywords metadata ATLEAST publisher

StanfordI I Ob jects Ob jectsself title author do cument pages for AND do ctoralthesisref FILLS

keywords metadata ATLEAST publisher StanfordI I

StanfordI Ob jects Ob jectsself title author NULL numb erofpages for AND do ctoralthesis ATLEAST

publisher StanfordI

WN Ob jects Ob jectsself name creator NULL pages for AND printmedia FILLS content thesis

content do ctoral FILLS generaltopics metadata WN

LSDIS Ob jects Ob jectsself title authors lo cationdo cument NULL forAND publications FILLS

typ e do ctoral thesis FILLS sub ject METADATA LSDIS

Based on the combination of partial translations the data retrieved from the rep ositories underlying

the ontologies can b e combined as follows

User Query Ob jects StanfordI I Ob jects
WN Ob jects

StanfordI Ob jects  LSDIS Ob jects

Using hyp onyms and hyp ernyms to interop erate across ontologies

Synonym relationships b etween terms in indep endent develop ed ontologies are very infrequent On

the contrary and real examples conrm it hierarchical relationships like hyponyms and hypernyms

are found more frequently The substitution of a term byitshyp ernyms or hyp onyms changes the

semantics of the queryWe try to translate the nontranslated terms in the user ontology into terms

which are not its synonyms in a target comp onentontology

We substitute a nontranslated term bytheintersection of its immediate parents or the union

of its immediate children The loss of information is measured in b oth cases and translation with

less loss of information is chosen This metho d is applied recursively until a full translation of the

conicting term is obtained Using hyp onym and hyp ernym relationships as describ ed ab ovecan

result in several p ossible translations of a nontranslated term into a target ontologyVery simple

intuitive measures dep ending on the extensions of the terms in the underlying ontologies may help

in cho osing the translations and minimizing the loss of information

In order to obtain the immediate parents and children of a term in the target ontologytwo

dierent kinds of relationships related to the conicting term must b e used

Synonyms hyp onyms and hyp ernyms b etween terms in the user and target ontology

Synonyms hyp onyms and hyp ernyms in the user ontology

The rst three typ es of relationships are stored in the IRM rep ository The second are relation

ships b etween terms in the same ontology synonyms are equivalent terms hyp onyms are those terms

subsumed by the nontranslated term and hyp ernyms are those terms that subsume the conicting

term

The task of getting the immediate parentschildren is not easy to p erform To obtain the

parentschildren within the user ontology the corresp onding functions eg subsumption of the

DL systems can b e used But wemust combine that answer with the immediate parentschildren in

the target ontologyTaking into account that some relationships stored in the IRM can b e redundant

they were indep endently dened by dierentontologies administrators such a task can b e quite

dicult Wewould need a DL system dealing with distributed ontologies

In Figure weshowtwoontologies with some relationships b etween them arrows are hyp onyms

relationships double arrows are synonyms and dashed lines are interontology relationships and

with the integrated ontology synonyms are group ed into one term on the right We can see that

obtaining the immediate parents is not evident for instance to get the immediate parents of B we

must deduce that A is a child of B There are also redundant relationships like the one b etween

A and B

Towork with the ab ove relationships in a homogeneous way an approachistointegrate the user

and the target ontologies and to use the deductivepower of the DL system to obtain the immediate

parentschildren of a term in the target ontology BIGP The prop erties b etween terms in the

dierentontologies are exactly the interontology relationships stored in the IRM so no intervention

of the user is needed Although some of the previous relationships can b e redundant the DL system

will classify the terms in the right place in the ontologyToknow if the resulting terms of the

integrated ontology are primitive or dened dep ending on A and B the rules describ ed in BIG

can b e used

Conclusions

Wehave discussed in this chapter the implications of the exp onential growth of the information on

the GI I on the semantic heterogeneity problem and explored new techniques to enable a solution A3 A1 A3 A1

B1 B1

A2#B3 INTEGRATION A2 B3 A1 B2 A1

B4 A4 A5 A4 A5 B4 B2

A6 A6

Figure Integrating twoontologies

to the same Information overload which arises as a consequence of the heterogeneity of the digital

data and media typ es is identied as the rst problem We explore an approach whereby metadata

descriptions are used to abstract out the representational details and characterize the information

content An informal classication of the various typ es of metadata used to handle the wide variety

of digital data was presented in Section The amount of information contentcapturedbyeach

is identied and domain specic metadata are identied as critical to the semantic heterogeneity

problem

We then discuss how approaches dep endent on representational or structural comp onents are in

adequate and argue the need for representation of contextual expressions in Section We discussed

the representation of these expressions using description logics and prop ose op erations to reason

with contextual expressions Weshowhow extra information whichmay not b e represented in the

database schema may b e represented using contextual descriptions We illustrated howcontextual

expressions may b e constructed from domain sp ecic ontologies and how terminological relationships

between concepts in an ontology enable representation of extra information

Wehave recognized the problem of vocabulary sharing as the most critical problem in con

struction of contextual descriptions We prop ose approaches to tackle the semantic heterogeneity

as opp osed to representational heterogeneity at this level in Section Semantic interop erability

across ontologies is enabled by utilizing terminological relationships like synonyms hypernyms and

hyponyms

Wehavethus explored various approaches based on metadata context and ontologies which

we b elieve are imp ortant and provide the required capabilities to handle the semantic heterogeneity

problem in the context of the GI I This research is a part of the InfoQuilt pro ject within the theme of

Enabling Infocosm KSb Fer at the Large Scale Distributed Information Systems Lab oratory

httplsdiscsugaedu at the University of Georgia Some of the interesting research topics that

are b eing investigated further in this this theme are as follows

Use of domain specic metadata to enable correlation of information across image and struc

tured data A future extension of this pro ject will b e to lo ok into use of metadata standards

suchasFGDC OGIS and domain sp ecic ontologies to describ e multimedia data

Extending the OBSERVER system to enable supp ort for hyponyms and hypernyms

Measures to characterize the loss of information accrued when a term is replaced by expres

sions with diering semantics These measures are b eing develop ed and exp erimented within

extended OBSERVER system

Providing a metadatabased reference link
A MREF  as an alternative to the physical

reference link
A HREF  This is b eing implemented as an extension to HTML on the

WWW SK This enables the publisher of an HTML do cument to sp ecify domain sp ecic

metadata which are then mapp ed to the underlying multimedia data by the enhanced server

This would enable a higherlevel metadata based metastructure over the currentWWW

References

ACHK Y Arens C Chee C Hsu and C Knoblo ck Retrieving and Integrating Data from

Multiple Information Sources International Journal of Intel ligent and Cooperative In

formation Systems June

AS J Anderson and M Stonebraker Sequoia Metadata Schema for Satellite Images

in KSc

BB A Borgida and R Brachman Loading Data into Description Reasoners In Proceedings

of ACM SIGMODMay

BBMR A Borgida R Brachman D McGuinness and L Resnick CLASSIC A structural data

mo del for ob jects In Proceedings of ACM SIGMOD

BIG JM Blanco A Illarramendi and A Goni Building a Federated Database System An

approach using a Knowledge Based System International Journal on Intel ligent and

Cooperative Information Systems Decemb er

BIGP J Blanco A Illarramendi A Goni and J Perez Using a terminological system to

integrate relational databases Information Systems Design and Hypermedia Cepadues

Editions

BL T BernersLee et al WorldWide Web The Information Universe Electronic Network

ing Research Applications and Policy

BR K Bohm and T Rakow Metadata for Multimedia Do cumentsinKSc

CHK F Chen M Hearst J Kupiec J Pederson and L Wilcox Metadata for MixedMedia

Access in KSc

CHS C Collet M Huhns and W Shen Resource Integration using a Large Knowledge Base

in Carnot IEEE Computer Decemb er

DDF S Deerwester S Dumais G Furnas T Landauer and R Hashman Indexing by

Latent Semantic Indexing Journal of the American Society for Information Science

DH U Dayal and H Hwang View denition and Generalization for Database Integration of

a Multidatabase System IEEE Transactions on Software Engineering November

Fer C Ferguson Into the info cosm Computerworld Leadership Series July

FKN PFankhauser M Kracker and E Neuhold Semantic vs Structural resemblance of

Classes SIGMOD Record special issue on Semantic Issues in Multidatabases A Sheth

ed Decemb er

Gru T Grub er Theory BIBLIOGRAPHICDATA September httpwww

kslstanfordeduknowledgesharingontologieshtmlbibliographicdataindexhtml

GSW U Glavitsch PSchauble and M Wechsler Metadata for Integrating Sp eech Do cuments

in a Text Retrieval System in KSc

HM D Heimbigner and D McLeo d A federated architecture for Information Systems ACM

Transactions on Oce Information Systems

HM J Hammer and D McLeo d An approach to resolving Semantic Heterogeneityina

Federation of Autonomous Heterogeneous Database Systems International Journal of

Intel ligent and Cooperative Information SystemsMarch

JH R Jain and A Hampapuram Representations of Video Databases in KSc

KKH Y Kiyoki T Kitagawa and T Hayama A metadatabase System for Semantic Im

age Searchby a Mathematical Mo del of Meaning SIGMOD Record special issue on

Metadata for Digital Media W Klaus A Sheth eds December

KM B Kahle and A Medlar An Information System for Corp orate Users Wide Area

Information Servers Connexions The Interoperability Report November

KSa V Kashyap and A Sheth Semanticsbased Information Brokering In Proceedingsofthe

Third International Conference on Information and Know ledge Management CIKM

Novemb er

KSb V Kashyap and A Sheth Semanticsbased Information Brokering A step towards

realizingthe Info cosm Technical Rep ort DCSTR Department of Computer Science

Rutgers UniversityMarch

KSc W Klaus and A Sheth Metadata for digital media SIGMOD Record special issue on

Metadata for Digital Media W Klaus A Sheth eds December

KS V Kashyap and A Sheth Schematic and Semantic Similarities b etween Database Ob

jects A Contextbased Approach Technical Rep ort TRCS LSDIS Lab Univer

sity of Georgia January Available at httplsdiscsugaeduamitcontext

algebraps An abridged version KS app ears in the VLDB Journal

KS V Kashyap and A Sheth Semantic and Schematic Similarities b etween Databases

Ob jects A Contextbased approach The VLDB Journal Octob er To

app ear httpwwwcsugaeduLSDISamitbVLDBps

KSS V Kashyap K Shah and A Sheth Metadata for building the MultiMedia Patch Quilt

In S Ja jo dia and V Subrahmanian editors MultiMedia Database Systems Issues and

Research Directions Springer Verlag

LA W Litwin and A Ab dellatif Multidatabase Interop erability IEEE Computer

Decemb er

LG D Lenat and R V Guha Building Large Know ledge Based Systems Representation

and InferenceintheCycProject AddisonWesley Publishing Company Inc

LNE J Larson S Navathe and R Elmasri A Theory of Attribute Equivalence in Databases

with Application to Schema Integration IEEE Transactions on Software Engineering

McC J McCarthy Notes on formalizing Context In Proceedings of the International Joint

ConferenceonArticial Intel ligence

Mil G Miller WordNet A Lexical Database for English Communications of the ACM

November

MKIS E Mena V Kashyap A Illarramendi and A Sheth Managing Multiple Information

Sources through Ontologies Relationship b etween Vo cabulary Heterogeneity and Loss

of Information In Proceedings of the workshop on Know ledge Representation meets

Databases in conjunction with European ConferenceonArticial Intel ligence August

MKSI E Mena V Kashyap A Sheth and A Illarramendi OBSERVER An approach

for query pro cessing in global information systems based on interop eration across pre

existing ontologies In Proceedings of the First IFCIS International ConferenceonCo

operative Information Systems CoopIS June

MS D McLeo d and A Si The Design and Exp erimental Evaluation of an Information

Discovery Mechanism for Networks of Autonomous Database Systems In Proceedings

of the th IEEE Conference on Data EngineeringFebruary

OM J Ordille and B Miller Distributed Active Catalogs and MetaData Caching in Descrip

tive Name Services In Proceedings of the th International Conference on Distributed

Computing SystemsMay

SG A Sheth and S Gala Attribute relationships An imp ediment in automating Schema

Integration In Proceedings of the NSF Workshop on Heterogeneous Databases December

She A Sheth Federated Database Systems for managing Distributed Heterogeneous and

Autonomous Databases Tutorial Notes the th VLDB ConferenceSeptemb er

Sho Y Shoham Varieties of Context

SK A Sheth and V Kashyap So Far Schematically yet So Near Semantically Invited

paper in Proceedings of the IFIP TCWG Conference on Semantics of Interoperable

Database Systems DSNovemb er In IFIP Transactions A North Holland

SK A Sheth and V Kashyap Mediaindep endent Correlation of Information What

How In Proceedings of the First IEEE Metadata Conference April

httplsdiscsugaedukashyapIEEEpap er

SL A Sheth and J Larson Federated Database Systems for managing Distributed Het

erogeneous and Autonomous Databases ACM Computing Surveys September

SLS K Sho ens A Luniewski PSchwartz J Stamos and J Thomas The Rufus System

Information Organization for SemiStructured Data In Proceedings of the th VLDB

Conference Septemb er

SSKS L Shklar A Sheth V Kashyap and K Shah Infoharness Use of Automatically Gen

erated Metadata for Search and Retrieval of Heterogeneous Information In Proceedings

of CAiSE June Lecture Notes in Computer Science

SSR E Sciore M Siegel and A Rosenthal Context Interchange using MetaAttributes In

Proceedings of the CIKM

Wie G Wiederhold Interop eration Mediation and Ontologies In FGCS Workshop on

Heterogeneous Cooperative Know ledgeBases Decemb er

A The LSDIS ontology

PUBLICATIONS

SUBJECT-BASED TYPE-BASED

WORKFLOW-PUB CONSISTENCY-PUB JOURNALS THESIS

INFORMATION-MODELING-PUB INTEGRATION-PUB TECHNICAL-REPORTS CONFERENCES

METADATA-PUB

B WN A subset of the WordNet ontology

PRINT-MEDIA

PUBLICATION JOURNALISM PRESS

FLEET-STREET WIRE-SERVICE NEWSPAPER MAGAZINE PHOTOJOURNALISM

DAILY PULP-MAGAZINE COMIC-BOOK

SLICK-MAGAZINE BOOK PERIODICAL

PICTORIAL SERIES JOURNALS TRADE-BOOK BROCHURE TEXTBOOK

MONTHLY WEEKLY QUATERLY BEST-SELLER TICKET-BOOK CRAMMER PRIMER REFERENCE-BOOK SONGBOOK

PRAYER-BOOK

BREVIARY MISSAL DIRECTORY COOKBOOK ENCYCLOPEDIA BOOK-OF-PSALMS PHONE-BOOK BLUE-BOOK INSTRUCTION-BOOK HANDBOOK ANNUAL WORDBOOK ALMANAC FARMERS-CALENDAR BIBLE GUIDEBOOK DICTIONARY THESAURUS MANUAL BILINGUAL-DICTIONARY POCKET-DICTIONARY ROADBOOK TRAVEL-BOOK

ETYMOLOGICAL-DICTIONARY INSTRUCTIONS REFERENCE-MANUAL

C StanfordI A subset of the Bibliographic Data ontology

BIBLIO-THING

AGENT CONFERENCE DOCUMENT

PERSON AUTHOR ORGANIZATION

PUBLISHER UNIVERSITY TECHNICAL-REPORT PROCEEDINGS

MISCELLANEOUS-PUBLICATION BOOK PERIODICAL-PUBLICATION

TECHNICAL-MANUAL MULTIMEDIA-DOCUMENT EDITED-BOOK JOURNAL NEWSPAPER THESIS COMPUTER-PROGRAM CARTOGRAPHIC-MAP MAGAZINE ARTWORK

MASTER-THESIS DOCTORAL-THESIS

D StanfordI I A subset of the Bibliographic Data ontology

REFERENCE

PUBLICATION-REF NON-PUBLICATION-REF

PERSONAL-COMMUNICATION-REF GENERIC-UNPUBLISHED-REF

BOOK-REF

TECHNICAL-REPORT-REF EDITED-BOOK-REF

PROCEEDINGS-PAPER-REF BOOK-SECTION-REF MISC-PUBLICATION-REF

ARTICLE-REF TECHNICAL-MANUAL-REF MULTIMEDIA-DOCUMENT-REF

JOURNAL-ARTICLE-REF NEWSPAPER-ARTICLE-REF COMPUTER-PROGRAM-REF CARTOGRAPHIC-MAP-REF THESIS-REF MAGAZINE-ARTICLE-REF ARTWORK-REF

DOCTORAL-THESIS-REF MASTER-THESIS-REF