1 1 Using Bayesian Decision for Ontology Mapping

1UsingBayesianDecisionforOntologyMappingJieTang*,JuanziLi,BangyongLiang,XiaotongHuang,YiLi,andKehongWang(DepartmentofComputerScienceandTechnology,TsinghuaUniversity,P.R.China,100084){j-tang02,liangby97,yi-li}@mails.tsinghua.edu.cn,{ljz,x.huang,wkh}@keg.cs.tsinghua.edu.cnAbstract.Ontologymappingisthekeypointtoreachinteroperabilityoverontologies.Insemanticwebenvironment,ontologiesareusuallydistributedandheterogeneousandthusitisnecessarytofindthemappingbetweenthembeforeprocessingacrossthem.Manyeffortshavebeenconductedtoautomatethediscoveryofontologymapping.However,someproblemsarestillevident.Inthispaper,ontologymappingisformalizedasaproblemofdecisionmaking.Inthisway,discoveryofoptimalmappingiscastasfindingthedecisionwithminimalrisk.AnapproachcalledRiMOM(RiskMinimizationbasedOntologyMapping)isproposed,whichautomatestheprocessofdiscoverieson1:1,n:1,1:nullandnull:1mappings.BasedonthetechniquesofnormalizationandNLP,theproblemofinstanceheterogeneityinontologymappingisresolvedtoacertainextent.Todealwiththeproblemofnameconflictinmappingprocess,weusethesaurusandstatisticaltechnique.Experimentalresultsindicatethattheproposedmethodcansignificantlyoutperformthebaselinemethods,andalsoobtainsimprovementovertheexistingmethods.Keywords.OntologyMapping,SemanticWeb,BayesianDecision,OntologyInteroperabilityContactinformationofCorrespondingAuthorName:TangJieE-mail:[e-mail]-Address:12#109,TsinghuaUniversity,Beijing,P.R.China100084Telephone:[phone]Fax:[phone]UsingBayesianDecisionforOntologyMappingJieTang*,JuanziLi,BangyongLiang,XiaotongHuang,YiLi,andKehongWang(DepartmentofComputerScienceandTechnology,TsinghuaUniversity,P.R.China,100084){j-tang02,liangby97,yi-li}@mails.tsinghua.edu.cn,{ljz,x.huang,wkh}@keg.cs.tsinghua.edu.cnAbstract.Ontologymappingisthekeypointtoreachinteroperabilityoverontologies.Insemanticwebenvironment,ontologiesareusuallydistributedandheterogeneousandthusitisnecessarytofindthemappingbetweenthembeforeprocessingacrossthem.Manyeffortshavebeenconductedtoautomatethediscoveryofontologymapping.However,someproblemsarestillevident.Inthispaper,ontologymappingisformalizedasaproblemofdecisionmaking.Inthisway,discoveryofoptimalmappingiscastasfindingthedecisionwithminimalrisk.AnapproachcalledRiMOM(RiskMinimizationbasedOntologyMapping)isproposed,whichautomatestheprocessofdiscoverieson1:1,n:1,1:nullandnull:1mappings.BasedonthetechniquesofnormalizationandNLP,theproblemofinstanceheterogeneityinontologymappingisresolvedtoacertainextent.Todealwiththeproblemofnameconflictinmappingprocess,weusethesaurusandstatisticaltechnique.Experimentalresultsindicatethattheproposedmethodcansignificantlyoutperformthebaselinemethods,andalsoobtainsimprovementovertheexistingmethods.Keywords.OntologyMapping,SemanticWeb,BayesianDecision,OntologyInteroperability1IntroductionOntologies,asthemeansforconceptualizingdomainknowledge,havebecomethebackbonetoenablethefulfillmentoftheSemanticWebvision[3].Manyontologieshavebeendefinedtomakedatasharable,forexample,CycOntology[17],EnterpriseOntology[38],Bibliographic-dataOntology[14],BiologicalandChemicalOntology(BAO)[25],andBio-Ontologies[43].See[45]formoreontologies.Unfortunately,ontologiesthemselvesaredistributedandheterogeneous.Ontologieshavetwokindsofheterogeneities:metadataheterogeneityandinstanceheterogeneity[4,16].Specifically,entities(entityrepresentsconcept,relation,orinstance)withthesamemeaningindifferentontologiesmayhavedifferentlabelnamesandthesamelabelnamemaybeusedforentitiesthathavedifferentintentionalmeanings;instancesindifferentontologiesmayhavedifferentrepresentations;anddifferentontologiesmayhavedifferenttaxonomystructures.Inordertoachievesemanticinteroperabilityoverontologies,itisnecessarytodiscover1SupportedbytheNationalNaturalScienceFoundationofChinaunderGrantNo.60443002*CorrespondingauthorE-mailaddresses:[e-mail],[e-mail],[e-mail],[e-mail],[e-mail],[e-mail],thefollowingproblemsstillexist.First,thenumberofcardinalitiesthatcanbeprocessedislimited.Mostoftheworkfocusesononly1:1mapping[9,10,15,18,22,23,26,29]despiteofthefactthatapproximately22%-50%ofmappingsarebeyondthiscardinalitybystatisticonreal-worldexamples[11,32].Secondly,Ontologymappingworkhasbeendonemainlyonmetadataheterogeneity,notoninstanceheterogeneity.Innaturallanguageprocessing,textnormalizationhasbeenstudied[34].Butbeforeadaptingthemethodologytoproblemofinstanceheterogeneity,manyeffortsarestillrequired.Theexistingmethodologiesproposedinthepreviousworkcanbeusedinontologymapping.However,theyarenotsufficientforsolvingalltheproblems.Atpresent,questionsariseforontologymapping:(1)howtoformalizetheproblemsothatitcandescribedifferentkindsofmappingcardinalitiesandheterogeneities,(2)howtosolvetheprobleminaprincipledapproach,and(3)howtomakeanimplementat