OMEN: A Probabilistic Ontology Mapping Tool

Download OMEN: A Probabilistic Ontology Mapping Tool

Post on 23-Jan-2016




0 download


OMEN: A Probabilistic Ontology Mapping Tool. Mitra et al. Mapping of two different ontologies. The Problem. We need to map databases or ontologies. The Problem. Mapping is difficult Most mapping tools are imprecise Even experts could be uncertain We deal with probabilistic mappings. - PowerPoint PPT Presentation


  • OMEN: A Probabilistic Ontology Mapping ToolMitra et al.

  • The ProblemWe need to map databases or ontologies

  • The ProblemMapping is difficultMost mapping tools are impreciseEven experts could be uncertainWe deal with probabilistic mappings

  • The SolutionInfer mappings based on previous onesWe use Bayesian Nets for inferenceWe use other tools for initial distributionsPreliminary results are encouraging

  • Basic ConceptsBayesian network:Probabilistic graphical model that represents Random variables

    Evidence nodes: The value is given

  • Bayesian NetworkConditional Probability tables (CPT)

  • Bayesian Nets in our approachHow do we build the Bayesian NetNodes are property or class matchesClasses are conceptsProperties are attributes of classes

  • Building Bayesian Nets

  • Our Bayesian NetsAll combinations of nodes is too manyWe generate only useful nodes The cutoff is k from evidence nodesUp to 10 parents per nodeCycles are avoided (confidence ~.5)

  • Our Bayesian NetsWe need evidence nodes and CPTsEvidence nodes come from initializationCPTs come from Meta-rules

  • Meta-rulesDescribes how other rules should be usedBasic Meta-rule

  • Other Meta-rulesRange: Restriction of property valuesMappings between properties and ranges of propertiesSingle rangeSpecialization

  • Other Meta-rulesMappings between super classesChildren matching depends on parents matchingFixed Influence Method (FI): P=.9Initial Probability Method (AP): P= y+cParent Probability Method (PP): P= x+c

  • Probability Distribution

  • Combining InfluencesWe assume that the parents are conditionally independentP[C|A,B] = P[C|A] x P[C|B]Fix of this for future work

  • Results2 Sets of 11 and 19 nodesPredicate matching was manualThresholds were .85 and .15

  • Results

  • StrengthsInnovative researchPublished at ISWCMathematically oriented

  • WeaknessesLots of typosNo comparison with current methodsLittle literature researchCould use better explanation of basic concepts

  • Future WorkHandling conditionally dependency of parent nodesHandling of matching predicatesAutomatic pruning and building of the network

  • ?