Open Research OnlineThe Open University’s repository of research publicationsand other research outputs

DOOR: towards a formalization of ontology relationsConference or Workshop ItemHow to cite:

Allocca, Carlo; d’Aquin, Mathieu and Motta, Enrico (2009). DOOR: towards a formalization of ontology relations. In:International Conference on Knowledge Engineering and Ontology Development (KEOD), 06-08 Oct 2009, Madera,Portugal.

For guidance on citations see FAQs.

c© Not known

Version: Accepted Manuscript

Link(s) to article on publisher’s website:http://www.keod.ic3k.org/Abstracts/2009/KEOD 2009 Abstracts.htm

Copyright and Moral Rights for the articles on this site are retained by the individual authors and/or other copyrightowners. For more information on Open Research Online’s data policy on reuse of materials please consult the policiespage.

oro.open.ac.uk

CORE Metadata, citation and similar papers at core.ac.uk

Provided by Open Research Online

DOOR:Towards a Formalization of Ontology Relations

Carlo Allocca, Mathieu d’Aquin and Enrico MottaKnowledge Media Institute, The Open University, Walton Hall, Milton Keynes, MK7 6AA, United Kingdom

c.allocca@open.ac.uk, M.Daquin@open.ac.uk, e.motta@open.ac.uk

Keywords: Ontology Relations, Networked Ontology

Abstract: In this paper, we describe our ongoing effort in describing and formalizing semantic relations that link ontolo-gies with each others on the Semantic Web in order to create an ontology, DOOR, to represent, manipulate andreason upon these relations. DOOR is a Descriptive Ontology of Ontology Relations which intends to definerelations such as inclusion, versioning, similarity and agreement using ontological primitives as well as rules.Here, we provide a detailed description of the methodology used to design the DOOR ontology, as well asan overview of its content. We also describe how DOOR is used in a complete framework (called KANNEL)for detecting and managing semantic relations between ontologies in large ontology repositories. Applied inthe context of a large collection of automatically crawled ontologies, DOOR and KANNEL provide a startingpoint for analyzing the underlying structure of the network of ontologies that is the Semantic Web.

1 INTRODUCTION

Ontologies are the pillars of the Semantic Web(SW) and, as more and more ontologies are madeavailable online, the SW is quickly taking shape. Asa result, the research community is becoming moreand more aware that ontologies are not isolated arti-facts: they are, explicitly or implicitly, related witheach other (Kleshchev and Artemjeva, 2005). Indeed,a number of studies have intended to tackle some ofthe challenges raised by these ontology relationships,from both theoretical and practical points of view.

At a theoretical level, studies have targetedontology comparison in order to identify overlapsbetween ontologies (Maedche and Staab, 2002).Approaches have been proposed to find differencesbetween versions of an ontologies (Noy and Musen,2002; Konev et al., 2008). According to (Kleinand Fensel, 2001), the ontology versioning problemhas been defined as the ability to handle changesin ontologies by creating and managing differentvariants of it. In other words, ontology versioningmeans that there are multiple variants of an ontologyaround. The authors of (Klein and Fensel, 2001)suggested that, ideally, developers should maintain

not only the different versions of an ontology, butalso some information about the way versions differand whether or not they are compatible with eachother. In (Gangemi et al., 1999) ontology integrationis defined as the construction of an ontology C thatformally specifies the union of the vocabularies oftwo other ontologies A and B. The most interestingcase is when A and B are supposed to commit to theconceptualization of the same domain of interest orof two overlapping domains. In particular, A and Bmay be related by being alternative ontologies, trulyoverlapping ontologies, equivalent ontologies withvocabulary mismatches, overlapping ontologies withdisjoint domain, homonymically overlapping ontolo-gies. Finally, in ontology matching, an alignment isset of correspondences between the entities of twoontologies, therefore relating these two ontologies bymapping there models with each other.At a practical level, Semantic Web Applications usethe SW as a large-scale knowledge source (d’Aquinet al., 2008): they achieve their tasks by automaticallyretrieving and exploiting knowledge from the SWas a whole, using advanced Semantic Web SearchEngines (SWSEs) such as WATSON (d’Aquinet al., 2007). These SWSEs provide keyword based

search mechanisms to locate relevant ontologies forparticular applications. As an example, the query“student” currently gives 1079 ontologies as a resultin WATSON1 (valid on the 22/04/2009). However,these results are provided as a simple list withoutmaking explicit the underlying relations that link on-tologies with each other. Indeed, on the first page, atleast 2 of the ontologies (http://www.vistology.com/ont/tests/student1.owl and http://www.vistology.com/ont/tests/student2.owl) repre-sent, apart from their URIs and the base namespaces,exactly the same logical model, expressed in thesame ontology language. Another common situationis when an ontology has been translated in differentontology languages. This is the case in the firstand second results of the query “student, university,researcher”(http://reliant.teknowledge.com/DAML/Mid-level-ontology.owl andhttp://reliant.teknowledge.com/DAML/Mid-level-ontology.daml). These two ontologiesare obviously two different encodings of the samemodel. Inspecting the results of WATSON in thesame way, it is not hard to find ontologies connectedwith other, more sophisticated semantic relations:versioning, inclusion, similarity, etc. Leaving im-plicit these relations in SWSE’s ontology repositoriesgenerates additional difficulties in exploiting theirresults, expecting the users and the applications tofind the “right” or “best” ontology to achieve theirgoal.

Both the theoretical and practical challenges con-cerning relations between ontologies indicate a needfor a general study of these relations, providing a for-mal base for defining, manipulating and reasoningupon the links that relate ontologies online, explic-itly or implicitly. Here, we chose to take an ontolog-ical approach to this problem. We design DOOR, aDescriptive Ontology of Ontology Relations that de-fines ontology relations using ontological primitivesand rules. Apart from the ontology itself, the maincontributions of this work concern the realization of amethodology to identify and define relations betweenontologies, as well as the development of a completesystem based on DOOR (KANNEL), providing ser-vices for detecting relations, populating DOOR, andformally querying detected and inferred relations in alarge ontology repository.

This paper is structured as follows: in Section 2we continue discussing significant work concerningontology relations; Section 3 presents the adoptedmethodology for designing DOOR; Section 4 de-scribes the DOOR ontology; In Section 5 we brieflydescribe KANNEL and the main role of DOOR in this

1http://watson.kmi.open.ac.uk

framework. Finally, Section 6 concludes the paperand sheds the light on interesting future research onontology relations.

2 RELATED WORK

J. Heflin (Heflin and Pan., 2004) was the first tostudied formally some of the different types of linksbetween ontologies, focusing on the crucial prob-lems of versioning and evolution. However, currently,there is no Ontology Management Systems that im-plements his framework. The authors of (Kleshchevand Artemjeva, 2005) characterized, at a very ab-stract level, a number of relations between ontolo-gies such as sameConceptualization, Resemblance,Simplification and Composition, without providingformal definitions for them, and without consider-ing the links between these relationships. Severalapproaches have been focusing on how to comparetwo different versions of ontologies in order to findthe differences. In particular, PROMTDIF (Noyand Musen, 2002) compares the structure of on-tologies and OWLDiff (http://semanticweb.org/wiki/OWLDiff) computes the differences by entail-ment, checking the two set of axioms. SemVer-sion (Volkel, 2006) compares two ontologies andcomputes the differences at both the structural andthe semantic levels. In addition, many measures ex-ist to compute particular forms of similarity betweenontologies (David and Euzenat, 2008).

All these studies discuss particular relations sepa-rately and are generally based on an abstract, informaldefinition of the relations they consider. A completemodel is necessary to provide a wide overview of ex-isting ontology relations, to clearly establish what aretheir definitions, formal properties, and how they areconnected with each other.

3 METHODOLOGY FOR THEDOOR ONTOLOGY

Building an ontology of relationships between on-tologies is a very ambitious task. It requires a deepanalysis of the ontologies available online and of theliterature, at different levels. Therefore, a reason-ably rigorous but nonetheless flexible methodology isneeded to identify, describe and formalize ontologyrelations and their connections, in order to build theDOOR ontology. Here, after defining some impor-tant elements that will be used in the rest of the paper,we present the steps involved in the methodology weadopted and briefly detail each of them.

3.1 Definitions and Requirements

We consider the following definitions:

Definition 1 (Ontology) An ontology is a set of ax-ioms (in the sense of the description logic) over a Vo-cabulary VOC, where VOC is the set of the primitiveterms (named entities) employed in the axioms of theontology;

Definition 2 (Ontology Space) An ontology spaceOS, is a collection of ontologies.

Definition 3 (Ontology Relation) Given an ontol-ogy space OS, an Ontology Relation is any binary re-lation defined over OS.

At the most general level, the design of the DOORontology was based on three main sources to identifyrelevant ontology relations:

1. We analyzed the results of SWSEs (e.g., WAT-SON) to manually identify existing, implicit re-lations between ontologies in these results.

2. We considered relations described in the litera-ture, such as the ones already mentioned in theprevious sections.

3. We also included existing, explicit relations thatare primitives of the OWL ontology language.

Also, ontology relations in the DOOR ontologyshould reflect the following important features:

• they are general enough to be applied to multipledomains;

• they are sufficiently intuitive to reflect generalmeaning;

• they are formally defined to be processed auto-matically by inference engines;

3.2 Main steps of the Methodology

To design DOOR, we considered the methodology de-scribed in (Gangemi et al., 2001) for selecting generalontological categories and adapted it to the problemof ontology relations. As a result, we divided our ap-proach into a number of steps, as follows:

1. Identifying the top level relations between ontolo-gies, considering our three sources (SWSEs, liter-ature and existing OWL primitives). At this stage,the task only consists in coming up with a list ofrelations that should be relevant, giving us a gen-eral overview of the different sections of the on-tology. Relations such as inclusion, similarity,incompatibility and previous version are identi-fied here.

2. Specifying the identified relations, identifying rel-evant variants and sub-relations. Here, our threesources of relations are also employed to deriverelations at a lower level than the top ones. Wealso use a more systematic approach, which con-sists in looking at ontologies (and so ontology re-lations) from 5 different dimensions that can char-acterize them:

• The Lexicographic level, which concerns thecomparison of the vocabularies of the ontolo-gies.

• Syntactic level, which concerns the compari-son of the sets of axioms that form the ontolo-gies.

• Structural level, which concerns the compari-son of the graph structure formed by the axiomsof the ontologies.

• Semantic level, which concerns the compar-ison of the formal models of the ontologies,looking in particular at their logical conse-quences.

• Temporal level, which concerns the analysis ofthe evolution of ontologies in time.

For example, considering the relation of inclusionidentified in the first step and that led to aproperty includedIn in the ontology, we canspecify this relation according to three dif-ferent dimensions (syntactic, structural andsemantic), leading to three variants of inclusionbetween ontologies (syntacticallyIncludedIn,isHomomorphicTo and semanticallyIncludedIn)that consider the set of axioms, the graph andthe formal models of the ontologies respectivly.In addition, besides the systematic analysis ofthis relation according to the dimensions, weinclude in DOOR particular forms of inclusionsderived from existing OWL primitives (e.g.,OWL imports) and from the literature (e.g,isAConservativeExtensionO f (Ghilardi et al.,2006)). More details about these relations aregiven in the next section.

3. Characterizing each relation by its algebraic prop-erties. For example, the algebraic properties forsimilarity are that it is re f lexive and symmetric.For inclusion, we can define that it is re f lexiveand transitive. Including such information in theontology corresponds to what (Gangemi et al.,2001) calls defining the ground axioms.

4. Establishing connections between relations. Theresults obtained from the previous steps aremainly top-level relations with a list of variants,each of them being given algebraic properties.

Here, we want to structure these lists, in partic-ular by giving them taxonomic relations. As anexample, it can be easily established that syn-tacticallySimilarTo is a sub property of semanti-callySimilarTo. In the same way, we can indicatethat a previous version of an ontology ought to besimilar to it. This corresponds to defining non-ground axioms in (Gangemi et al., 2001).

5. Introducing rules to define complex relations fromatomic ones. For example, the equivalentToproperty can be defined as equivalentTo(X1, X2):-includedIn(X1, X2), includedIn(X2, X1).

Like in any methodology, the application of thesesteps should be flexible and continuous. Getting backto a previous step is sometimes necessary and, as thebuilding of an ontology such as DOOR is a constantlyongoing effort, it should be possible to re-apply themethodology entirely to make the ontology evolve.

The intended result is an ontology made, on theone hand, of an ontologically defined and taxonomi-cally structured set of relations, and on the other hand,of a set of rules to define complex relations. In thefollowing we give a detailed overview of the first ver-sion of the DOOR ontology, considering only the first(ontological) part of it, as, due to its complexity, thedefinition of rules governing complex relations is stilla work in progress and would not fit in this paper.

4 FORMAL DESCRIPTION OFDOOR

The OWL version of the DOOR ontology canbe downloaded at: http://kannel.open.ac.uk/ontology.We start with describing the first level ofDOOR, in Figure 2. The main relevant abstract re-lations are simply represented as sub-properties ofontologyRelatedTo. An ontology X is ontologyRe-latedTo another one Y if one of the top level re-lations is satisfied between X and Y. The top levelrelations include includedIn, similarTo, isAlignedTo,disagreesWith, agreesWith and isTheSchemaFor. Weclustered them in four groups and each group will beexplained in more details in the next sub-sections.

4.1 includedIn and equivalentTo

includedIn and equivalentTo are two of the main on-tology relations. The former represents the meaningof “an ontology contains an another one”. The lat-ter intends to convey the meaning of “two ontolo-gies express the same knowledge”. According to ourmethodology, these two relations have been analyzedat different levels, giving origin to different kinds of

ontologyRelatedTo

agreesWith disagreesWith includedIn isAlignedTo isTheSchemaFor similarTo

Figure 1: Top Level of DOOR.

inclusion and equivalence relations. In Table 1, wesummarize the result of these analyses:

Table 1: Specialization of inclusion and equivalence rela-tions.

includedIn equivalentTo

Semantic semanticallyIncludedIn semanticallyEquivalentTo

isAConservativeExtentionOf

Structural isHomomorphicTo isIsomorphicTo

Syntactic syntacticallyIncludedIn syntacticallyEquivalentTo

import

In particular, the sub-relations of includedIn are de-fined as follows:

SyntacticallyIncludedIn(X1, X2) if the set of axiomsof X1 is contained in the set of axioms of X2,whichmeans X1 ⊆ X2.

isHomomorphicTo(X1, X2) if a homomorphism ex-ists between the RDF-graph of X1 and the RDF-graph of the X2.

SemanticallyIncludedIn(X1, X2) if the set of modelsof X1 is contained in the set of models of X2. Inother words, if X1 |= X2.

isAConservativeExtentionOf(X1, X2) , informally, ifsyntacticallyIncludedIn(X2,X1) and all the ax-ioms entailed by X1 over the vocabulary of X2 arealso entailed by X2. A more formal definition canbe found in (Ghilardi et al., 2006). The notion ofconservative extension has been used in particularfor ontology modularization (Grau et al., 2007).

import(X1, X2) if there is an explicit statement in X1indicating that it imports X2 using the owl:importsprimitive. Formally, this means that all the axiomsof X2 should be considered as contained in X1.

The sub-relations of equivalentTo are defined as fol-lows:

syntacticallyEquivalentTo(X1, X2) if and only ifSyntacticallyIncludedIn(X1, X2) and Syntac-ticallyIncludedIn(X2, X1).

syntacticallyEquivalentTo

equivalentTo isHomomorphicTosemanticallyIncludedIn

includedIn

isIsomorphicTo semanticallyEquivalentTo

imports isAConservativeExtensionOf

isIsomorphicTosyntacticallyIncludedIn

Figure 2: Taxonomy for includedIn and equivalentTo

isIsomorphicTo(X1, X2) if an isomorphism exists be-tween the graph of X1 and the graph of X2.

SemanticallyEquivalentTo if and only ifSemanticallyIncludedIn(X1, X2) andSemanticallyIncludedIn(X2, X1).

Finally, following our methodology, we defined thealgebraic properties of each relation2 and classifiedthem to create a taxonomic structure relating these re-lations. This structure is showed in Figure 23.

4.2 similarToOntology similarity has been described as a measureto assess how close two ontologies are (David andEuzenat, 2008). Various ways to compute the sim-ilarity between two ontologies have been describedwhich are relevant in different application contexts. Inour work, similarTo is used to represent the meaningof “how an ontology overlap/cover parts of the samearea of interest of another ontology”. Following ourmethodology, similarTo has been analyzed and for-malized at the lexicographic, structural and semanticlevel, giving origin to different kinds of similarity re-lations (see Table 2).

To define these relations, we need to introduce thefollowing elements: given two ontologies X1 and X2,we denote by LC(X1, X2) the set of axioms of X1 thatare logical consequences of X2 and by Voc(X1) the vo-cabulary of X1. The following definitions depend ona threshold T > 0.

2Since these are fairly obvious, we do not detail then.3The arrows represent the subPropertyOf relation. For

example, syntacticallyEquivalentTo is a sub property of se-manticallyIncludedIn.

Table 2: Specialization of the similarity relation

SimilarTo

Semantic SemanticallySimilarTo

MappingSimilarTo

Syntactic SyntatticallySimilarTo

Lexicographic LexicographicSimilarTo

owlIncompatibleWithexplanationEvolution syntacticallyEquivalentTo

equivalentTo isLatterVersionOf isPreviousVersionOf

similarTo

isIsomorphicTo sematicallyEquivalentTo syntacticallySimilarTopriorVersionbackwardCompatibleWith conceptualEvolutionOf explanationEvolution

lexicallySimilarTo mappingSimilarTo semanticallySimilarTo

Figure 3: Taxonomy for similarTo. Dashed elements rep-resent elements from other sections of the ontology.

semanticallySimilarTo(X1, X2), if

|LC(X1,X2)T

LC(X2,X1)|max(|X1|, |X2|)

≥ T

syntacticallySimilarTo(X1, X2), if

|X1T

X2|max(|X1|, |X2|)

≥ T

lexicographicallySimilarTo(X1, X2), if|Voc(X1)

TVoc(X2)|

max(|Voc(X1)|, |Voc(X2)|)≥ T

Finally, in addition to the relations defined above,we also consider a similarity relation that relies on theexistence of an alignment between the two ontologies.Indeed, mappingSimilarTo is a relation that linkstwo ontologies X1 and X2 if there exists an alignmentfrom X1 to X2 and this alignment covers a substantialpart of the vocabulary of X1 (i.e., a proportion greaterthan a threshold T ). Not that, since alignments canbe unidirectional, mappingSimilarTo differs fromthe other similarity functions by not being symmetric.

Finally, we have classified the relations in Table 1to create the taxonomic structure showed in Figure 3.

4.3 VersioningVersioning is a temporal relation that concerns theevolution of an ontology. In (Klein and Fensel, 2001),

the ontology versioning problem has been defined asthe ability to handle changes in ontologies by creatingand managing different variants of it.

An ontology can evolve over time in different di-rections, e.g. lexicographic, changing the names ofsome resources, syntactic, adding or removing ax-ioms, semantic, changing the definition of some con-cepts or simply adding or removing axioms. There-fore, the new ontology could be equivalent or totallydifferent from the previous one. When we analyzedifferent ways of linking two ontologies by the ver-sioning relation, the two following sentences are sug-gested immediately: “X1 is the previous version ofthe X2” or “X2 is the latter version of the X1”. Thesetwo typical pieces of knowledge are represented in theDOOR ontology by the relations isPreviousVersionOfand its inverse isLatterVersionOf respectively.

Conforming to our methodology, the isPrevi-ousVersionOf and isLatterVersionOf relations havebeen analyzed and formalized, to identify sub-relations and variants. In Table 3 we summarize theresult of this analysis.

Table 3: Specialization of the versioning relations.

isLatterVersionOf isPreviousVersionOf

Temporal conceptualEvolutionOf priorVersion

explanationEvolutionOf

backwardCompatibleWith

owl:IncompatibleWith

Semantic conceptualEvolutionOf

Syntactic explanationEvolutionOf

According to (Klein et al., 2002; Heflin and Pan.,2004; Heflin, 2001) the modification of an ontologycan lead to two different types of evolutions: beinga conceptual change, meaning that the model of theontology changed, or being an explanation change,meaning that the modifications happened only at asyntactic level, without affecting the model of theontology. Therefore, we specialized the isLatterVer-sionOf relation into

conceptualEvolutionOf(X1,X2) if X1 is a latter ver-sion that is not semantically equivalent to X2.

explanationEvolutionOf(X1,X2) if X1 is a latter ver-sion that is semantically equivalent to X2

These two relations will lead to the definition of rulesto infer them from equivalence and other versioningrelations.

In addition, the OWL ontology propertiespriorVersion, backwardCompatibleWith andincompatibleWith represent explicit relations be-tween versions of ontologies and are included in

owlIncompatibleWithexplanationEvolution syntacticallyEquivalentTo

equivalentTo isLatterVersionOf isPreviousVersionOf

similarTo

isIsomorphicTo sematicallyEquivalentTo syntacticallySimilarTopriorVersionbackwardCompatibleWith conceptualEvolutionOf explanationEvolution

lexicallySimilarTo mappingSimilarTo semanticallySimilarTo

Figure 4: Taxonomy for versioning relations.

DOOR as sub-properties of isLatterVersionO f andisPreviousVersionO f .

To complete this section of the DOOR ontology,we can classified the relations in Table 3 as showed inFigure 4.

Indeed, according to (Klein et al., 2002; Heflinand Pan., 2004; Heflin, 2001) the modification of anontology can lead a new version which is completelydifferent from the original one. But in practice, by an-alyzing Watson’s ontology repository, it is almost al-ways possible establish a similarity between the twoontologies, at least at the lexicographic level. Due tothis fact, we chose to consider the versioning relationsto be sub-properties of similarTo, to indicate that twodifferent versions of the same ontology should, tosome extent, be similar. Moreover, in accordance withits definition, the explanationEvolutionOf relation isa sub-property of semanticallyEquivalentTo.

4.4 Agree and DisagreeBased on the formal measure of the agreement anddisagreement between two ontologies presented in(d’Aquin, 2009), we introduce the agreesWith anddisagreesWith relations in DOOR. Informally, the for-mer holds the general meaning of “to have the sameopinion about something”. In other words, it connectstwo ontologies, sharing the same knowledge partiallyand is therefore very related to the similarTo and theequivalentTo relations. The later indicates that the on-tologies “contradict each other” to a certain extent,these contradictions appearing at various levels. En-visaged sub-relations for these two relations are listedin Table 4.

In this Table, all the sub-relations of agreesWithhave already been defined before. We add a few re-lations to express specific ways for ontologies to dis-agree, all related to the semantic dimension of the on-

Table 4: Specialization of agreesWith and disagreesWith.

agreeWith disagreeWith

Temporal backwardCompatibleWith owlIcompatibleWith

Semantic sematicallyEquivalentTo hasDisparateModeling

sematicallySimilarTo incompatibleWith

incoherentWith

incosistentWith

Syntactic syntacticallyEquivalentTo

syntacticallySimilarTo

explanationEvolution

agreesWith

semanticallyEquivalentToBackwardCompatibleWith

explanationEvolution syntacticallyEquivalentTo syntactiallySimilarTo

semanticallySimilarTo

Figure 5: Taxonomy for the agreement relations.

tologies.

incompatibleWith(X1, X2) if incoherentWith(X1, X2)or inconsistentWith(X1, X2).

incoherentWith According to (Qi and Hunter, 2007)an ontology X1 is incoherent if and only if thereis an unsatisfiable concept name in X1. Therefore,two ontologies are incoherent with each other if,when they are merged, they generate an incoher-ent result.

inconsistentWith According to (Bell et al., 2007) anontology X1 is inconsistent if it has no model.Therefore, two ontologies are inconsistent witheach other if, when they are merged, they generatea inconsistent result.

hasDisparateModeling besides the logical notion ofdisagreement, we also consider disagreements re-lated to the conceptualization of the ontologies.Two ontologies are considered to have disparatemodeling if they represent corresponding entitiesin different ways, e.g. as an instance in one caseand a class in the other.

owl:CompatibleWith It comes from OWL language(Patel-Schneider et al., 2004).

Finally, we have also classified the relations in Ta-ble 4 as showed in Figures 5 and 6.

4.5 isTheSchemaFor and isAlignedToAnalyzing Watson’s ontology repository we found outthat there are many documents which only represent

hasDisparateModelling incompatibleWith

disagreeWith

incoherentWith inconsistentWith owlIncompatibleWith

Figure 6: Taxonomy for the disagreement relations.

the TBox of an ontology and others representing justthe ABox. By the isTheSchemaFor relation we wantto keep the link between the TBox and the corre-sponding ABox, which is very important when wewant to retrieve information.isAlignedTo relation isnot well defined yet.

5 KANNEL: AN APPLICATIONFOR THE DOOR ONTOLOGY

In the previous section, we described the DOORontology in detail. Here we provide a brief overviewof the way it is used in the KANNEL system. KAN-NEL (Allocca., 2009) is a framework for detectingand managing ontology relations for large ontologyrepositories, such as WATSON. It is an ontology-based system where the DOOR ontology plays an im-portant role, providing an explicit representation ofthe implicit relations between ontologies. We havedesigned an architecture for this framework, as de-picted in Figure 7. As showed in this figure, theDOOR Ontology separates the on-line part of thearchitecture–providing APIs and services that relieson a reasoner–from the off-line part–detecting rela-tions in the repository and populating the ontology.The offline part is based on three components: theControl Component (CC), the Detecting Component(DC) and the Populating Component (PC). As a firststep, the CC selects from the Ontology Repository on-tologies that need to be evaluated to establish poten-tial relations. Then, the selected sets of ontologies areprocessed by the DC, which contains the main mech-anisms to discover the possible relations between on-tologies, relying on the definitions provided in this pa-per. Finally, the PC populates the semantic structurewith the detected relations. What is obtained is a setof automatically discovered relations, represented aspart of the DOOR ontology so that the reasoner usedin the system can infer new relations from the onto-logical and rule based knowledge included in the on-tology. As such, DOOR provides meta-informationon the ontology repository in KANNEL.

Figure 7: Architecture of the KANNEL framework.

6 CONCLUSIONSIn this paper, general relationships between on-

tologies have been examined. In particular, we havechosen to consider well-known relations in the litera-ture, as well as the ones needed to support the devel-opment of Semantic Web Applications. To achievethat, we adapted an ontology building methodologyfor the construction of DOOR, an ontology of rela-tions between ontologies. This ontology describesrelations both from the point of view of their taxo-nomic structure and from the point of view of theirformal definitions, providing the formal properties todescribe them as well as a set of rules to derive com-plex relations from other relations.

We also described KANNEL, a framework for de-tecting and managing ontology relationships for largeontology repositories. The DOOR ontology plays afundamental role in KANNEL, not only to provide anexplicit representation on ontology relations, but alsoto supply meta-information that offers several advan-tages, among which the possibility to reason upon on-tologies and their relations. This possibility providesa relevant support for the development of SemanticWeb Applications, which can use the semantic web asa large-scale knowledge source (d’Aquin et al., 2008).

The first version of the DOOR ontology isavailable in OWL at http://kannel.open.ac.uk/ontology. The KANNEL framework is currently un-der development. The development of DOOR is ob-viously a continuous task, which requires a proper as-sessment of each version. For this reason, we planto test and validate the first version presented here,in particular by populating it with automatically de-tected relations between ontologies in WATSON.

ACKNOWLEDGEMENTSThis work was funded by the EC IST-FF6-027595

NeOn Project.

REFERENCES

Allocca., C. (2009). Expliciting semantic relations betweenontologies in large ontology repositories. PhD Sym-posium, Poster Session, ESWC.

Bell, D., Qi, G., and Liu, W. (2007). Approaches to incon-sistency handling in description-logic based ontolo-gies. Proc of the SWWS Conference., 4825/2008.

d’Aquin, M. (2009). Formally measuring agreement anddisagreement in ontologies. 5th K-CAP.

d’Aquin, M., Motta, E., and et al, M. S. (2008). Towardsa new generation of semantic web applications. IEEEIntell. Sys., 23(3).

d’Aquin, M., Sabou, M., Dzbor, M., Baldassarre, C.,Gridinoc, L., Angeletou, S., and Mottta, E. (2007).Watson: A gateway for the semantic web. Poster Ses-sion at 4th ESWC.

David, J. and Euzenat, J. (2008). Comparison between on-tology distances (preliminary results). 7th Int. Seman-tic Web Conference, ISWC.

Gangemi, A., Guarino, N., and Masolo, C. (2001). Under-standing top-level ontological distinctions.

Gangemi, A., Pisanelli, D. M., and Steve, G. (1999). Anoverview of the onions project: Applying ontologiesto the integration of medical terminologies. Technicalreport. ITBM-CNR, V. Marx 15, 00137, Roma, Italy.

Ghilardi, S., Lutz, C., and Wolter, F. (2006). Did I damagemy ontology? a case for conservative extensions indescription logics. In 10th Inter. Conf. (KR), pages187–197. AAAI Press.

Grau, B. C., Horrocks, I., Kazakov, Y., and Sattler, U.(2007). Just the right amount: Extracting modulesfrom ontologies. In Proceedings of WWW, pages 717–726, Banff, Canada. ACM.

Heflin, J. (2001). Towards the semantic web: Knowledgerepresentation in a dynamic, distributed environment.Ph.D. Thesis, University of Maryland, 2001.

Heflin, J. and Pan., Z. (2004). A model theoretic semanticsfor ontology versioning. 3th ISWC, Hiroshima, Japan,LNCS 3298 Springer, pages 62–76.

Klein, M. and Fensel, D. (2001). Ontology versioning onthe semantic web. Proc. of the Inter. Semantic WebWorking Symposium (SWWS), pages 75–91.

Klein, M., Fensel, D., Kiryakov, A., and Ognyanov, D.(2002). Ontology versioning and change detection onthe web. 13th Intern Conf on Know. Engineering andKnow. Management (EKAW02), pages 197–212.

Kleshchev, A. and Artemjeva, I. (2005). An analysis ofsome relations among domain ontologies. Int. Journalon Inf. Theories and Appl, 12:85–93.

Konev, B., C.Lutz, D.Walther, and F.Wolter (2008). Cex andmex: Logical diff and logic-based module extractionin a fragment of owl. Liverpool University, UK andTU Dresden, Germany.

Maedche, A. and Staab, S. (2002). Comparing ontologies-similarity measures and a comparison study. Proc. ofEKAW-2002.

Noy, N. F. and Musen, M. A. (2002). Promptdiff: A fixed-point algorithm for comparing ontology versions. 18thNational Conf. on Artificial Intelligence (AAAI).

Patel-Schneider, P. F., Hayes, P., and Horrocks, I. (2004).Owl web ontology language semantics and abstractsyntax. W3C Recommendation.

Qi, G. and Hunter, A. (2007). Measuring incoherence indescription logic-based ontologies. Proc of the Int.Sem. Web Conference., 4825/2008:381–394.

Volkel, M. (2006). D2.3.3.v2 SemVersion Versioning RDFand Ontologies. EU-IST Network of Excellence (NoE)IST-2004-507482 KWEB.