Your SlideShare is downloading. ×
0
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Ontologies and Similarity
Upcoming SlideShare
Loading in...5
×

Thanks for flagging this SlideShare!

Oops! An error has occurred.

×
Saving this for later? Get the SlideShare app to save on your phone or tablet. Read anywhere, anytime – even offline.
Text the download link to your phone
Standard text messaging rates apply

Ontologies and Similarity

348

Published on

Keynote Talk at Int Conf on Case-based Reasoning

Keynote Talk at Int Conf on Case-based Reasoning

Published in: Technology, Education, Spiritual
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total Views
348
On Slideshare
0
From Embeds
0
Number of Embeds
0
Actions
Shares
0
Downloads
23
Comments
0
Likes
0
Embeds 0
No embeds

Report content
Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
No notes for slide

Transcript

  • 1. Ontologies and Similarity<br />Steffen Staab<br />Acknowledgements to Claudia d’Amato, Univ Bari, <br />&amp; WeST Team<br />TexPoint fonts used in EMF. <br />Read the TexPoint manual before you delete this box.: AAAAAAAAAAA<br />
  • 2. Agenda<br />Kris: Brocoliisvegetableused in stirfry<br />Motivation Whatareexamplesemanticapplications?<br />FoundationWhatis an ontology?<br />Reality Check Whataretypicalontologies?<br />Survey Howissimilaritymeasured in ontologies? <br />CritiqueWhatshouldbemeasured?<br />Solution A preliminarysolution<br />ConclusionWhatto do now?<br />
  • 3. Motivation<br />SemanticApplications<br />Check out: http://challenge.semanticweb.org/<br />
  • 4. Linked Data<br />Cases withMetadatawithout Frontiers<br />
  • 5. Semantic Search &amp; Browsing: Semantic Portals<br />[WWW 2000]<br />http://ontoprise.com<br />
  • 6. FacetedSemantic Media Browsing: Semaplorer<br />Winner Billion Triples Challenge 2008 [JoWS 2009]<br />http://kreuzverweis.com<br />
  • 7. Semantic Desktop<br />Additional Semantic Meta Data, e.g. sender, subject<br />Access to further PIM tools<br />
  • 8. Mobile Exploration ofLinked Data: Mobile Facets<br />
  • 9. LessonsLearned<br />Examples + http://challenge.semanticweb.org<br />Semantic Boolean Search in Conjunction with Keyword Search dominates in <br /><ul><li>Ontology-based applications
  • 10. Linked data applications</li></ul>Feast or famine<br />Further useofsimilarity<br /><ul><li>Learning
  • 11. Ontologyengineeringadvice</li></ul>Available<br /><ul><li>IR Ranking
  • 12. (Textual) Similarity</li></ul>Needed<br /><ul><li>Semantic Ranking
  • 13. SemanticSimilarity</li></ul>[Franz et al 09]<br />[stuffhere], BUT<br />
  • 14. Whatis an Ontology?<br />Foundation<br />
  • 15. Whatis an ontology?<br />Whatfor?<br />Agreements thatmakelinkeddatamoreuseful<br />Reasoning<br />Gruber 1993: <br />An ontology is an “explicit specificationof a conceptualization”<br />Oberle, Guarino, Staab. Whatis an ontology? Handbook on ontologies, Springer 2009.<br />
  • 16. Observations in the Real World<br />
  • 17. A Model ofthe Real World<br />knows<br />knows<br />Manager(I034820)<br />Researcher(I046758)<br />knows<br />cooperates<br />Employee(I050000)<br />Researcher(I044443)<br />
  • 18. Abstractingfromthe Individual Model<br />knows<br />knows<br />Manager<br />Researcher<br />knows<br />cooperates<br />Employee<br />Researcher<br />
  • 19. A Conceptual Model<br />Intensional Relations<br />Unary<br />Manager<br /> Research <br />Employee<br />Binary<br />cooperates<br />knows<br />Cognitive Bias<br />Perception<br />Knowledge<br />Belief<br />The conceptualmodelcaptureswhatis invariant accordingtoone‘sconceptualizationoftheworld<br />
  • 20. Formal Specification<br />Whatmakesit so hardtoformallyspecifyontologicalcommitment?<br />Algebraic Relations do not work: <br /><ul><li>Definedextensionally
  • 21. E.g. Lecturer1 = {Ashwin, Nirmalie, Steffen, Kris,…}
  • 22. Problem: New instancewouldchangetheontology, e.g.Lecturer2 = Lecturer1  {Fernando}</li></ul>Intensional Relations needtobedefined in Higher Order Language:<br /><ul><li>Specifytheintendedmodelswhereonemayquantifyoversetsofindividuals</li></ul>An ontologyis a theory (typically in firstorderlogicallanguage) wherethepossiblemodelsapproximatetheintendedmodels „asgoodaspossible“<br />
  • 23. Conceptualization<br />Perception<br />Reality<br />State of affairs<br />State of affairs<br />relevant invariants across presentation patterns:D, <br />Presentationpatterns<br />Phenomena<br />Bad <br />Ontology<br />Ontological commitmentK (selects D’D and ’)<br />Models MD’(L)<br />Ontology<br />InterpretationsI<br />Intended models for each IK(L)<br />Ontology models<br />Language L<br />~Good<br />Slide by Nicola Guarino<br />
  • 24. Description Logics: First orderlanguage(s) forontology<br />T-Box<br />Describing Relations Intensionally<br />Flight  Service.<br />Flight  ∃to.Airport<br />Flight  to.Airport<br />Flight  ∃from.Airport<br />Flight  from.Airport<br />approachedBy ⊇ to-1<br />FlightFromDE = Flight ∩ ∃from.(Airport ∩part.{DE})<br />A-Box<br />Describing Relations Extensionally<br />Flight(LH123).<br />Flight(BA121).<br />Airport(FRA).<br />from(LH123,FRA).<br />to(LH123,LHR).<br />…<br />Key Feature: Classes (unaryrelations) aredefinedbyrelationstodefinitionsofotherclasses<br />
  • 25. Description Logics: First orderlanguage(s) forontology<br />T-Box<br />Describing Relations Intensionally<br />Flight  Service.<br />Flight  ∃to.Airport<br />Flight  to.Airport<br />Flight  ∃from.Airport<br />Flight  from.Airport<br />domain(to) ⊇ Flight<br />FlightFromDE = Flight ∩ ∃from.(Airport ∩part.{DE})<br />A-Box<br />Describing Relations Extensionally<br />Flight(LH123).<br />Flight(BA121).<br />Airport(FRA).<br />from(LH123,FRA).<br />to(LH123,LHR).<br />…<br /><ul><li>Typicallydecidableandintractable
  • 26. Pragmaticallytractablefor 105concepts
  • 27. Oftenmostusefulat design time only</li></li></ul><li>WhataretypicalOntologies?<br />Reality Check<br />
  • 28. ExamplesforOntologies &amp; Thesauri<br />Foundational Model ofAnatomy<br /><ul><li>78K classes in FMA 2.0
  • 29. Severaltranslationsto OWL fordiscoveringmodelingproblems ([Noy &amp; Rubin; Bodenreider et al])</li></ul>SNOMED CT(Systematized Nomenclature of Medicine -- Clinical Terms)<br /><ul><li>Representation in descriptionlogicslanguage EL++
  • 30. 106classes</li></ul>Dewey Decimal System<br /><ul><li>Internationallyusedthesaurusforformingpre-coordinatedclassesfrom an inventoryofcodes</li></li></ul><li>Examplefrom Dewey Decimal<br />590 Animals (Zoology)<br />770 Photography, Computer Art<br />597.96 Serpentes<br />779.32 Photographyof Animals<br />779.32796 PhotographyofSnakes<br />Core messageofthistalk:<br />Influencing also non-OWL ontologies/thesauri<br />Conceptsaredefinedbased on therelationshiptothedefinitionofotherconceptsaffectingsimilarity<br />
  • 31. Howissimilaritymeasured in ontologies?<br />Survey<br />
  • 32. ExampleOntology<br />Airport<br />Service<br />Europe<br />part<br />Hub<br />part<br />part<br />Flight<br />LHR<br />IT<br />UK<br />FCO<br />LCY<br />DE<br />FRA<br />part<br />part<br />Including „invariant“ A-Box facts(like Airport(FRA))<br />to<br />to<br />to<br />FRA-LCY<br />FRA-LHR<br />FRA-FCO<br />
  • 33. Similarity Measurement Tasks<br />ComparingClasses<br />Comparing Objects<br /><ul><li>Based on objectfeatures
  • 34. Based on classcomparisons</li></ul>ComparingOntologies<br /><ul><li>Lexeme comparisons
  • 35. Graph comparison
  • 36. Consideringthesemanticsofhierarchies
  • 37. isa
  • 38. part
  • 39. Other relations</li></ul>Relatedto<br /><ul><li>Ontologylearning
  • 40. Ontologyalignment</li></ul>Based on<br /><ul><li>Class comparisons</li></li></ul><li>Class Comparisons in MaterializedHierarchies<br />Airport<br />Service<br />Europe<br />part<br />part<br />part<br />Flight<br />LHR<br />IT<br />UK<br />FCO<br />LCY<br />DE<br />FRA<br />part<br />part<br />to<br />to<br />to<br />FRA-LCY<br />FRA-LHR<br />FRA-FCO<br />
  • 41. Class Comparisons in MaterializedHierarchies<br />Airport<br />Service<br />Europe<br />part<br />part<br />part<br />Flight<br />LHR<br />IT<br />UK<br />FCO<br />LCY<br />DE<br />FRA<br />part<br />part<br />Flight-DE-UK<br />Flight-DE-IT<br />Howmanyyellowconcepts?<br /><ul><li>Infinitelymanyin powerfulDL languages</li></ul>to<br />to<br />to<br />FRA-LCY<br />FRA-LHR<br />FRA-FCO<br />
  • 42. IntensionalCountingof Path Length<br />𝑠𝑖𝑚𝐹𝑅𝐴−𝐿𝐻𝑅,𝐹𝑅𝐴−𝐿𝐶𝑌 ~ 1𝑃𝑎𝑡h1=12<br /> <br />Service<br />𝑠𝑖𝑚𝐹𝑅𝐴−𝐿𝐻𝑅,𝐹𝑅𝐴−𝐹𝐶𝑂 ~ 1𝑃𝑎𝑡h2=14<br /> <br />Flight<br />3 importantobservations:<br /><ul><li>Most papersinvestigatedampening, i.e. higher links indicatemoredissimilarity
  • 43. Absolute similarityvaluesmostly irrelevant (like in CBR)
  • 44. Most information in theontology will bediscarded</li></ul>Flight-DE-UK<br />Flight-DE-IT<br />FRA-LCY<br />FRA-LHR<br />FRA-FCO<br />[Rada et al.&apos;89] ff<br />
  • 45. IntensionalCountingof Path Length<br />𝑠𝑖𝑚𝐹𝑅𝐴−𝐿𝐻𝑅,𝐹𝑅𝐴−𝐿𝐶𝑌 ~ 1𝑃𝑎𝑡h1=1min2,3=12<br /> <br />Service<br />𝑠𝑖𝑚𝐹𝑅𝐴−𝐿𝐻𝑅,𝐹𝑅𝐴−𝐹𝐶𝑂 ~ 1𝑃𝑎𝑡h2=1min4,2=12 <br /> <br />Flight<br />Flight-DE-UK<br />Flight-DE-IT<br />FlightFromHub<br />FlightToHub<br />FlightFrom+ToHub<br />FRA-LCY<br />FRA-LHR<br />FRA-FCO<br />
  • 46. `Improved´IntensionalCountingof Path Length<br /> 𝑐𝑜𝑡𝑜𝑝𝑦𝐶=𝐷 | 𝑖𝑠𝑎∗(𝐶,𝐷)<br /> <br />Service<br />𝑠𝑖𝑚𝑐𝑜𝑡𝑜𝑝𝑦(𝐶,𝐷)~ |𝑐𝑜𝑡𝑜𝑝𝑦(𝐶) ∩𝑐𝑜𝑡𝑜𝑝𝑦(𝐷)||𝑐𝑜𝑡𝑜𝑝𝑦(𝐶)⋃𝑐𝑜𝑡𝑜𝑝𝑦(𝐷)|<br /> <br />𝑠𝑖𝑚𝑐𝑜𝑡𝑜𝑝𝑦𝐹𝑅𝐴−𝐿𝐻𝑅,𝐹𝑅𝐴−𝐹𝐶𝑂 ~ 59<br /> <br />Flight<br />Further dampeningpossible<br />Flight-DE-UK<br />Flight-DE-IT<br />FlightFromHub<br />FlightToHub<br />FlightFrom+ToHub<br />FRA-LCY<br />FRA-LHR<br />FRA-FCO<br />
  • 47. `Improved´ IntensionalCountingof Path Length - Jaccard<br /> 𝑐𝑜𝑡𝑜𝑝𝑦𝐶=𝐷 | 𝑖𝑠𝑎∗(𝐶,𝐷)<br /> <br />Service<br />𝑠𝑖𝑚𝑐𝑜𝑡𝑜𝑝𝑦(𝐶,𝐷)~ |𝑐𝑜𝑡𝑜𝑝𝑦(𝐶) ∩𝑐𝑜𝑡𝑜𝑝𝑦(𝐷)||𝑐𝑜𝑡𝑜𝑝𝑦(𝐶)⋃𝑐𝑜𝑡𝑜𝑝𝑦(𝐷)|<br /> <br />𝑠𝑖𝑚𝑐𝑜𝑡𝑜𝑝𝑦𝐹𝑅𝐴−𝐿𝐻𝑅,𝐹𝑅𝐴−𝐹𝐶𝑂 ~ 59<br /> <br />Flight<br />𝑠𝑖𝑚𝑐𝑜𝑡𝑜𝑝𝑦𝐹𝑅𝐴−𝐿𝐻𝑅,𝐹𝑅𝐴−𝐿𝐶𝑌 ~ 48<br /> <br />Flight-DE-UK<br />Flight-DE-IT<br />FlightFromHub<br />FlightToHub<br />FlightFrom+ToHub<br />FRA-LCY<br />FRA-LHR<br />FRA-FCO<br />
  • 48. Intension basedSimilarity Measurement<br />Strengths<br />Works somehow<br />Weaknesses<br />Bothpathcounting/Cotopyheavilysufferfrommodellingartefacts in theontology<br />
  • 49. CountingExtensions – Jaccard-likeMetrics<br />𝑠𝑖𝑚𝐹𝑅𝐴−𝐿𝐻𝑅,𝐹𝑙𝑖𝑔h𝑡𝑇𝑜𝐻𝑢𝑏 ~ |𝐹𝑅𝐴−𝐿𝐻𝑅 ∩𝐹𝑙𝑖𝑔h𝑡𝑇𝑜𝐻𝑢𝑏||𝐹𝑅𝐴−𝐿𝐻𝑅 𝐹𝑙𝑖𝑔h𝑡𝑇𝑜𝐻𝑢𝑏|=36<br /> <br />Service<br />𝑠𝑖𝑚𝐹𝑅𝐴−𝐿𝐻𝑅,𝐹𝑅𝐴−𝐿𝐶𝑌 ~ |𝐹𝑅𝐴−𝐿𝐻𝑅 ∩𝐹𝑅𝐴−𝐿𝐶𝑌||𝐹𝑅𝐴−𝐿𝐻𝑅 𝐹𝑅𝐴−𝐿𝐶𝑌|=04<br /> <br />Flight<br />Flight-DE-UK<br />Flight-DE-IT<br />FlightFromHub<br />FlightToHub<br />FlightFrom+ToHub<br />FRA-LCY<br />FRA-LHR<br />FRA-FCO<br />Disjointnessincompatibility<br />LH127<br />LH123<br />BA124<br />BA121<br />LH345<br />LH567<br />AI234<br />[Resnik ‘95-‘99]<br />
  • 50. Extension basedSimilarity<br />Strengths<br />Countingextensionsseemsnaturalandefficient(Jaccard-likemeasure)<br />Weaknesses<br />DisjointnessIncompatibility<br />Classesaresimilar, but do not shareinstances:<br /><ul><li>Male – Female
  • 51. Housecat – Lion</li></ul>Extensionsareuncountable<br />Ontologiessupposedtoabstractfromspecificextensions!<br />Extensionsmaybe infinite<br />
  • 52. Class Syntax basedSimilarity<br />Quitefrequent in theliterature<br />Listedhere just forsakeofcompleteness, because…<br />Class syntaxbasedsimilarityis<br />equivalenceunsound<br />
  • 53. WhatshouldSimilarityDeliver?<br />Critique<br />[d‘Amato et al 2008]<br />
  • 54. Core criteriaforsimilaritymeasures– almostunchanged<br />Positiveness: C,D sim(C,D)  0<br />Strong reflexivity:Csim(C,C) = 1<br />Upperbound: C,D sim(C,D)  1<br />Symmetry: C,D sim(C,D) = sim(D,C) <br />Problem with strong reflexivity:<br />FlightFromDEHub = Flight ∩∃from.(Hub ∩part.{DE}) FromHubAndFromDE = ∃from.Hub∩∃from.part.{DE}<br />Reasoningisneededtodiscoverthat<br />sim(FlightFromDEHub,FromHubAndFromDE) = 1<br />But problem:<br />FlightFromDEHub = Flight ∩∃from.(Hub ∩part.{DE}) FromHubAndFromDE = ∃from.Hub∩∃from.part.{DE}<br />Reasoningisneededtodiscoverthat<br />sim(FlightFromDEHub,FromHubAndFromDE) = 1<br />
  • 55. Additional Ones in Ontologies!<br />5. PreventDisjointnessIncompatibility (seenbefore)<br />6. Equivalence Soundness: <br /> C,D,E DE  sim(C,D)=sim(C,E) <br />Example:<br />sim(Flight,FlightFromDEHub) =<br />sim(Flight,FromHubAndFromDE)<br />Proposition:Reflexivityandtriangleinequalityimplyequivalencesoundness<br />
  • 56. Additional Ones in Ontologies!<br />7. Monotonicity<br />CL, DL, CU, DU,<br />EU, E⊆L<br />∃H such thatCH, EH, DH<br /> sim(C,D)  sim(C,E)<br />U<br />L<br />C<br />D<br />E<br />Myfeelingis: weneedmore!<br />(continuity,…)<br />
  • 57. A Preliminary Solution<br />Solution<br />[d‘Amato et al 2010]<br />
  • 58. Core idea: Combine Cotopy &amp; Extension-basedApproaches<br />Cotopy-based: IntersectionattheLeastCommonSubsumer<br />Extension-based: Count instances (orsubclasses)<br />Venndiagramsindicates: sim(C,D) &gt; sim(C,E)<br />E<br />gcs(C,D)<br />C<br />C<br />D<br />gcs(C,E)<br />
  • 59. Indirect (tentative) Indicationof Correctness<br />Growingindexingtreebyclusteringwithnewsimilaritymeasure<br />Comparingquerying time for different ontologiesusingthe original hierarchyandtheindexingtreederivedfromsimilaritymeasure<br />Problem: similaritycomputationtoo expensive<br />[d‘Amato et al 2010]<br />
  • 60. Whatto do now?<br />Conclusion<br />
  • 61. Conclusion<br />Conclusion: A call to arms!<br /><ul><li>Semanticapplicationscovermanydomainsofcommercialandsocialinterest
  • 62. Ontologiesprovidethemodelingbackboneandareevenfound in unexpectedplaces
  • 63. Similaritymeasuresforontologiesexistandgive back someresults
  • 64. Criteriaforsemanticsimilaritymeasuresare still in themaking
  • 65. Thereis a lack oftheoryforontology-basedsimilarity
  • 66. Thereis a lack ofefficientrealizationofontology-basedsimilarity</li></ul>Targeted Side Effect:<br />ClarificationofSomeOftenMistakenUseofTerminologyaroundOntologies<br />
  • 67. Institut WeST – Web Science &amp; Technologies<br />ThankYou!<br />Semantic Web<br />Web Retrieval<br />Interactive Web<br />Multimedia Web<br />Software Web<br />eGovernment<br />eMedia<br />eScience<br />eOrganizations<br />eCitizen<br />

×