Abstract

A survey on ontology mapping

Description
A survey on ontology mapping
Categories
Published
of 8
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
Share
Transcript
    A Survey on Ontology Mapping Namyoun Choi, Il-Yeol Song, and Hyoil HanCollege of Information Science and TechnologyDrexel University, Philadelphia, PA 19014 Abstract Ontology is increasingly seen as a key factor forenabling interoperability across heterogeneous systemsand semantic web applications. Ontology mapping isrequired for combining distributed and heterogeneousontologies. Developing such ontology mapping hasbeen a core issue of recent ontology research. Thispaper presents ontology mapping categories, describesthe characteristics of each category, compares thesecharacteristics, and surveys tools, systems, and relatedwork based on each category of ontology mapping. Webelieve this paper provides readers with acomprehensive understanding of ontology mapping andpoints to various research topics about the specific rolesof ontology mapping. Introduction “An ontology is defined as a formal, explicitspecification of a shared conceptualization.” 27    Taskson distributed and heterogeneous systems demandsupport from more than one ontology. Multipleontologies need to be accessed from different systems.The distributed nature of ontology development has ledto dissimilar ontologies for the same or overlappingdomains. Thus, various parties with different ontologiesdo not fully understand each other. To solve theseproblems, it is necessary to use ontology mappinggeared for interoperability. This article aims to presentthe broad scope of ontology mapping, mappingcategories, their characteristics, and a comprehensiveoverview of ontology mapping tools, systems, andrelated work.We classify ontology mapping into the followingthree categories: 1) mapping between an integratedglobal ontology and local ontologies 3, 4, 1, 7 , 2) mappingbetween local ontologies 6, 1, 8, 9, 12, 13, 14 , and 3) mappingon ontology merging and alignment. 15, 16, 17, 18, 19, 20 The first category of ontology mapping supportsontology integration by describing the relationshipbetween an integrated global ontology and localontologies. The second category enablesinteroperability for highly dynamic and distributedenvironments as mediation between distributed data insuch environments. The third category is used as a partof ontology merging or alignment as an ontology reuseprocess.In this paper, we survey the tools, systems, andrelated work about ontology mapping based on thesethree ontology mapping categories. A comparison of tools or systems about ontology mapping is madebased on specific evaluation criteria 10 , which areinput requirements, level of user interaction, type of output, content of output, and the following fivedimensions: structural, lexical, domain, instance-based knowledge, and type of result. 8 Through acomparative analysis of ontology mapping categories,we aim to provide readers with a comprehensiveunderstanding of ontology mapping and point tovarious research topics about the specific roles of ontology mapping.The paper is organized as follows. The meaningsof ontology mapping 4, 3, 7, 15, 25 , ontology integration,merging, and alignment 2, 24 are outlined in Section 2.In Section 3, characteristics and application domainsof three different categories of ontology mapping arediscussed. The tools, systems, frameworks, andrelated work of ontology mapping are surveyed basedon the three different ontology mapping categories.Then the overall comparison of tools or systemsabout ontology mapping is presented. In Section 4, aconclusion and presentation of future work aredetailed. 2. Terminology: ontology mapping, ontologyintegration, merging, and a lignment In this section, we set the scope of ontologymapping and ontology mapping tools, and outlinemeanings of ontology mapping, integration, merging,and alignment. We aim to give a wide view of ontology mapping including ontology integration,merging, and alignment because this concept of ontology mapping is broad in scope 5 and ontologymapping is required in the process of ontologyintegration, merging, and alignment. Furthermore,one closely related research topic with ontologymapping is schema matching, which has been onemajor area of database research. 3, 36, 37, 38 However,this is beyond our scope in this paper. We also referto tools for ontology integration, merging, andalignment as ontology mapping tools in this paper.We discuss the meanings of ontology mapping basedon the three different ontology mapping categories. Ontology merging, integration, and alignment Ontology merging, integration, and alignmentcan be considered as an ontology reuse process. 2,24   34SIGMOD Record, Vol. 35, No. 3, Sep. 2006  Ontology merging is the process of generating a single,coherent ontology from two or more existing anddifferent ontologies related to the same subject. 26 Amerged single coherent ontology includes informationfrom all source ontologies but is more or lessunchanged. The srcinal ontologies have similar oroverlapping domains but they are unique and notrevisions of the same ontology. 24  Ontology alignment is the task of creating linksbetween two srcinal ontologies. Ontology alignment ismade if the sources become consistent with each otherbut are kept separate. 15 Ontology alignment is madewhen they usually have complementary domains.Ontology integration is the process of generating asingle ontology in one subject from two or moreexisting and different ontologies in different subjects. 26  The different subjects of the different ontologies maybe related. Some change is expected in a singleintegrated ontology. 26 Ontology mappingOntology mapping between an integrated globalontology and local ontologies. 4, 3, 7 In this case,ontology mapping is used to map a concept found inone ontology into a view, or a query over otherontologies (e.g. over the global ontology in the local-centric approach, or over the local ontologies in theglobal-centric approach). Ontology mapping between local ontologies. 25 Inthis case, ontology mapping is the process thattransforms the source ontology entities into the targetontology entities based on semantic relation. The sourceand target are semantically related at a conceptual level. Ontology mapping in ontology merge andalignment. 15   In this case, ontology mapping establishescorrespondence among source (local) ontologies to bemerged or aligned, and determines the set of overlapping concepts, synonyms, or unique concepts tothose sources. 15 This mapping identifies similarities andconflicts between the various source (local) ontologiesto be merged or aligned. 5 3. Categories of Ontology Mapping In this section, ontology mapping based on thefollowing three categories will be examined: 1)ontology mapping between an integrated globalontology and local ontologies, 2) ontology mappingbetween local ontologies, and 3) ontology mapping inontology merging and alignment.One of the crucial differences among the threeontology mapping categories is how mapping amongontologies is constructed and maintained. Eachcategory of ontology mapping has differentcharacteristics (strengths and drawbacks). Ontologymapping plays an important role in differentapplication domains 5 and is the foundation of severalapplications. 14 3.1   Ontology mapping between an integratedglobal ontology and local ontologies This category supports ontology integrationprocesses. Methodological aspects of ontologyintegration relate to how this mapping is defined. 1  This mapping specifies how concepts in global andlocal ontologies map to each other, how they can beexpressed based on queries 7 , and how they aretypically modeled as views or queries (over themediated schema in the local-as-view approach, orover the source schemas in the global-as-viewapproach). 7   3.1.1   Strengths and drawbacks The strengths of this mapping can also be thedrawbacks of mapping between local ontologies andvice versa. In this mapping, it is easier to definemapping and find mapping rules than in mappingbetween local ontologies because an integrated globalontology provides a shared vocabulary and all localontologies are related to a global ontology. It can bedifficult to compare different local ontologiesbecause no direct mappings exist between localontologies. This mapping lacks maintainability andscalability because the change of local ontologies orthe addition and removal of local ontologies couldeasily affect other mappings to a global ontology.This mapping requires an integrated global ontology.But there exists a practical impossibility of maintaining it in a highly dynamic environment. 8  This mapping cannot be made among differentontologies which have mutually inconsistentinformation over the same domain or over a similarview of domain because a global ontology cannot becreated. 3.1.2   Application domains This mapping supports the integration of ontologies for the Semantic Web, enterpriseknowledge management, and data or informationintegration. In the Semantic Web, an integratedglobal ontology extracts information from the localones and provides a unified view through which userscan query different local ontologies. 7 Whenmanaging multiple ontologies for enterpriseknowledge management, different local ontologies(data sources) can be combined into an integratedglobal ontology for a query. 1 In an informationintegration system, a mediated schema is constructed SIGMOD Record, Vol. 35, No. 3, Sep. 200635  for user queries. Mappings are used to describe therelationship between the mediated schema (i.e., anintegrated global ontology) and local schemas. 1,7,3,4  Ontology is more complicated and expressive insemantics than schema and has some differences butshares many features. 34, 35, 5 Schema can still be viewedas an ontology with restricted relationship types. 9  Therefore, the mediated schema can be considered as aglobal ontology. 3  3.1.3   Tools, systems, and related work  An integrated global ontology (the logicalmediated schema) is created as a view. 4,7,3 Mappingsare used to describe the relationship between themediated schema and local schemas. LSD 3 (Learning Source Description): LSD semi-automatically creates semantic mappings with a multi-strategy learning approach. This approach employsmultiple learner modules with base learners and themeta-learner where each module exploits a differenttype of information in the source schemas or data. LSDuses the following base learners: 1) The Name Learner:it matches an XML element using its tag name, 2) TheContent Learner: it matches an XML element using itsdata value and works well on textual elements, 3) NaïveBayes Learner: it examines the data value of theinstance, and doesn’t work for short or numeric fields,and 4) The XML Learner: it handles the hierarchicalstructure of input instances. Multi-strategy learning hastwo phases: training and matching. In the training phase,a small set of data sources has been manually mappedto the mediated schema and is utilized to train the baselearners and the meta learner. In the matching phase,the trained learners predict mappings for new sourcesand match the schema of the new input source to themediated schema.   LSD also examines domain integrityconstraints, user feedback, and nested structures inXML data for improving matching accuracy. LSDproposes semantic mappings with a high degree of accuracy by using the multi-strategy learning approach. MOMIS 4 (Mediator Environment for MultipleInformation Sources): MOMIS creates a global virtualview (GVV) of information sources, independent of their location or their data’s heterogeneity. MOMISbuilds an ontology through five phases as follows:1)   Local source schema extraction by wrappers   2)   Local source annotation with the WordNet3)   Common thesaurus generation: relationshipsof inter-schema and intra-schema knowledgeabout classes and attributes of the sourceschemas4) GVV generation: A global schema and mappingsbetween the global attributes of the globalschema and source schema by using the commonthesaurus and the local schemas are generated.5)   GVV annotation is generated by exploitingannotated local schemas and mappingsbetween local schemas and a global schema.MOMIS generates mappings between globalattributes of the global schema and source schemas.For each global class in the global virtual view(GVV), a mapping table (MT) stores all generatedmappings. MOMIS builds an ontology that moreprecisely represents domains and provides an easilyunderstandable meaning to content, a way to extendpreviously created conceptualization by inserting anew source. A Framework for OIS 7 (Ontology IntegrationSystem): Mappings between an integrated globalontology and local ontologies are expressed asqueries and ontology as Description Logic. Twoapproaches for mappings are proposed as follows: 1)concepts of the global ontology are mapped intoqueries over the local ontologies (global-centricapproach), and 2) concepts of the local ontologies aremapped to queries over the global ontology (local-centric approach). 3.2   Ontology mapping between local ontologies This category provides interoperability forhighly dynamic, open, and distributed environmentsand can be used for mediation between distributeddata in such environments. 12 This mapping is moreappropriate and flexible for scaling up to the Webthan mappings between an integrated global ontologyand local ontologies. 12   3.2.1   Strengths and drawbacks This mapping enables ontologies to becontextualized because it keeps its content local. 6 Itcan provide interoperability between local ontologieswhen different local ontologies cannot be integratedor merged because of mutual inconsistency of theirinformation. 6,1 It is useful for highly dynamic, open,and distributed environments 6 and also avoids thecomplexity and overheads of integrating multiplesources. 1 Compared to mapping between anintegrated ontology and local ontologies, thiscategory mapping has more maintainability andscalability because the changes (adding, updating, orremoving) of local ontology could be done locallywithout regard to other mappings. Finding mappingsbetween local ontologies may not be easier thanbetween an integrated ontology and local ontologiesbecause of the lack of common vocabularies. 3.2.2   Application domains The primary application domains of thismapping are the Web or the Semantic Web because 36SIGMOD Record, Vol. 35, No. 3, Sep. 2006  of their de-centralized nature. When there is no centralmediated global ontology and coordination has to bemade using ontologies, then mappings between localontologies are necessary for agents to interoperate. 14 Indistributed knowledge management systems, whenbuilding an integrated view is not required or multipleontologies cannot be integrated or merged because of mutual inconsistency of the information sources, thiscategory of mapping is required between localontologies. 1,6   3.2.3   Tools, systems, and related workContext OWL 6 (Contextualizing Ontologies):OWL syntax and semantics are extended. Ontologiescannot be integrated or merged as a single ontology if two ontologies contain mutually inconsistent concepts.However, those two ontologies can be mapped usingbridge rules which are the basic notion about thedefinition of context mappings. 6 A mapping betweentwo ontologies is a set of bridge rules using ⊇ , ⊆ , ≡ , ∗  (related), and ⊥ (unrelated). CTXMATCH 8 :   CTXMATCH is an algorithm fordiscovering semantic mappings across hierarchicalclassifications (HCs) using logical deduction.CTXMATCH takes two inputs H, and H1 in HCs, andfor each pair of concepts k  ∈ H , k1 ∈ H1 (a node withrelevant knowledge including meaning in Hierarchicalclassifications), returns their semantic relation ( ⊇ , ⊆ , ≡ , ∗ , and ⊥ ). For example, k is more general than k1 (k  ⊇  k1), k is less general than k1 (k  ⊆ k1), k is equivalent tok1 (k  ≡ k1), k is compatible with k1 (k  ∗ k1), and k isincompatible with k1 (k  ⊥ k1).The contribution of the CTXMTCH is thatmappings can be assigned a clearly defined model-theoretic semantics and that structural, lexical, anddomain knowledge are considered. GLUE 9 : GLUE semi-automatically createsontology mapping using machine learning techniques.GLUE consists of Distribution Estimator, SimilarityEstimator, and Relaxation Labeler.   GLUE finds themost similar concepts between two ontologies andcalculates the joint probability distribution of theconcept using a multi-strategy learning approach forsimilarity measurement. GLUE gives a choice to usersfor several practical similarity measures. GLUE has atotal of three learners: Content Learner, Name Learner,and Meta Learner. Content and Name Learners are twobase learners, while Meta Learner combines the twobase learners’ prediction. The Content Learner exploitsthe frequencies of words in content of an instance(concatenation of attributes of an instance) and uses theNaïve Bayes’ theorem. The Name Learner uses the fullname of the input instance. The Meta-Learner combinesthe predictions of base learners and assigns weights tobase learners based on how much it trusts that learner’spredictions. In GLUE, Relaxation Labeling takes asimilarity matrix and reaches for the mapping (bestlabel assignment between nodes (concepts)). Thismapping configuration is the output of GLUE. MAFRA 12 (Ontology MA apping FRA mework for distributed ontologies in the Semantic Web):MAFRA provides a distributed mapping process thatconsists of five horizontal and four verticalmodules. 12 Five horizontal modules are as follows:1)   Lift & Normalization: It deals with languageand lexical heterogeneity between sourceand target ontology.2)   Similarity Discovery: It finds out andestablishes similarities between sourceontology entities and target ontology entities.3)   Semantic Bridging: It defines mapping fortransforming source instances into the mostsimilar target instances.4)   Execution: It transforms instances from thesource ontology into target ontologyaccording to the semantic bridges.5)   Post-processing: It takes the result of theexecution module to check and improve thequality of the transformation results.Four vertical modules are as follows:1)   Evolution: It maintains semantic bridges insynchrony with the changes in the sourceand target ontologies.2)   Cooperative Consensus Building: It isresponsible for establishing a consensus onsemantic bridges between two parties in themapping process.3)   Domain Constraints and BackgroundKnowledge: It improves similarity measureand semantic bridge by using WordNet ordomain-specific thesauri.4)   Graphical User Interface (GUI): Humanintervention for better mapping.MAFRA maps between entities in two differentontologies using a semantic bridge, which consists of concept and property bridges. The concept bridgetranslates source instances into target ones. Theproperty bridge transforms source instance propertiesinto target instance properties. LOM 21 (Lexicon-based Ontology Mapping):LOM finds the morphism between vocabularies inorder to reduce human labor in ontology mappingusing four methods: whole term, word constituent,synset, and type matching. LOM does not guaranteeaccuracy or correctness in mappings and haslimitations in dealing with abstract symbols or codesin chemistry, mathematics, or medicine. QOM 22 (Quick Ontology Mapping): QOM is aefficient method for identifying mappings betweentwo ontologies because it has lower run-timecomplexity. In order to lower run-time complexity SIGMOD Record, Vol. 35, No. 3, Sep. 200637  QOM uses a dynamic programming approach. 33 Adynamic programming approach has data structureswhich investigate the candidate mappings, classify thecandidate mappings into promising and less promisingpairs, and discard some of them entirely to gainefficiency. It allows for the ad-hoc mapping of large-size, light-weight ontologies. ONION 13   ( ON tology composit ION system):ONION resolves terminological heterogeneity inontologies and produces articulation rules for mappings.The linguistic matcher identifies all possible pairs of terms in ontologies and assigns a similarity score toeach pair. If the similarity score is above the threshold,then the match is accepted and an articulation rule isgenerated. After the matches generated by a linguisticmatcher are available, a structure-based matcher looksfor further matches. An inference-based matchergenerates matches based on rules available withontologies or any seed rules provided by experts.Multiple iterations are required for generating semanticmatches between ontologies. A human expert chooses,deletes, or modifies suggested matches using a GUItool. A linguistic matcher fails when semantics shouldbe considered. OKMS 1 (Ontology-based knowledge managementsystem): OKMS is an ontology-based knowledgemanagement system. In OKMS, mapping is used forcombining distributed and heterogeneous ontologies.When two different departments deal with the samebusiness objects, their ontologies for their systems donot match because they approach the domain fromdifferent perspective. When they want to includeinformation from other departments in their knowledgemanagement system, the information must betransformed (i.e., reclassified). This can beaccomplished through a mapping between localontologies. The five-step   ontology-mapping   process 12 isused in the OKMS. The five-step ontology mappingprocess is as follows: 1) Lift and normalization: If source information is not ontology-based, it will betransformed to the ontology level by a wrapper. 2)Similarity extraction: The similarity extraction phasecreates a similarity matrix, which represents thesimilarities between concepts and instances inontologies being mapped. 3) Semantic mapping: Thisstep produces the mappings that define how totransform source-ontology instances into target-ontology instances. 4) Execution: Execute mappings. 5)Post-processing: It improves the results of the executionphase. OMEN 31 (Ontology Mapping Enhancer): OMEN isa probabilistic ontology mapping tool which enhancesthe quality of existing ontology mappings using aBayesian Net. The Bayesian Net uses a set of meta-rules that represent how much each ontology mappingaffects other related mappings based on ontologystructure and the semantics of ontology relations.Existing mappings between two concepts can be usedfor inferring other mappings between relatedconcepts. P2P ontology mapping 32 : This work  32 proposesthe framework which allows agents to interact withother agents efficiently based on the dynamicmapping of only the portion of ontologies relevant tothe interaction. The framework executes three steps:1) Generates the hypotheses. 2) Filters the hypotheses.3) Selects the best hypothesis. 3.3   Ontology mapping (matching) in ontologymerging and alignment This category allows a single coherent mergedontology to be created through an ontology mergingprocess. It also creates links between local ontologieswhile they remain separate during the ontologyalignment process. Mappings do not exist between asingle coherent merged ontology and local ontologies,but rather between local ontologies to be merged oraligned. Defining a mapping between localontologies to be merged or aligned is the first step inthe ontology merging or alignment process. Thismapping identifies similarities and conflicts betweenlocal ontologies to be merged or aligned. 3.3.1   Strength and drawbacks This mapping applies to ontologies over thesame or overlapping domain. Finding mapping is apart of other applications such as ontology mergingor alignment. This might be fairly obvious and moreinteresting in a large ontology. 14,11   3.3.2   Application domains The growing usage of ontologies or thedistributed nature of ontology development has led toa large number of ontologies which have the same oroverlapping domains. 15,17 These should be merged oraligned to be reused. 15 Many applications such asstandard search, e-commerce, governmentintelligence, medicine, etc., have large-scaleontologies and require the reuse of ontology mergingprocesses. 11   3.3.3   Tools, systems, and related workSMART 18 : SMART is a semi-automaticontology merging and alignment tool. It looks forlinguistically similar class names through class-namematches, creates a list of initial linguistic similarity(synonym, shared substring, common suffix, andcommon prefix) based on class-name similarity, 38SIGMOD Record, Vol. 35, No. 3, Sep. 2006
Search
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks