1 / 42

OWL はオントロジーの課題を 本当に解決するのか？

人工知能学会研究会資料 SIG-SWO-A201-02. OWL はオントロジーの課題を 本当に解決するのか？. 山口　 高平 （静岡大学）. Semantic Web の９階層. 知識工学のトレンド 90-: 概念化の明示的仕様 （ Tom Gruber オントロジーの定義） オントロジー記述言語 (Ontolingua) 知識交換言語 (KIF) Generic Ontology CYC, WordNet, EDR … PSM Task Ontology オントロジー構築方法論 ．．．. オントロジーに関する 知識工学と次世代 Web の流れ.

eliana-maldonado
Download Presentation

OWL はオントロジーの課題を 本当に解決するのか？

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. 人工知能学会研究会資料SIG-SWO-A201-02 OWLはオントロジーの課題を本当に解決するのか？ 山口　 高平 （静岡大学）

  2. Semantic Web の９階層

  3. 知識工学のトレンド 90-: 概念化の明示的仕様（Tom Gruber オントロジーの定義）オントロジー記述言語(Ontolingua)知識交換言語(KIF)Generic OntologyCYC, WordNet, EDR…PSMTask Ontologyオントロジー構築方法論．．． オントロジーに関する知識工学と次世代Webの流れ 次世代Webのトレンド • 95-97: XML as arbitrary structures • 97-98: RDF • 98-99: RDFS (schema) as a frame-like system • 00-01: DAML+OIL • 02-07: OWL

  4. OWLへの道のり Jan’00 Med’00 Sep’00 Dec’00 Mar’01 Aug’01 Oct’01 • On-To-Knowledge が OILを定義 • 重要EUプロジェクトのいくつかが OILを採用 • DAMLプロジェクトとOILがリンク • 定義： DAML+OIL • 改定： DAML+OIL • W３Cへ DAML+OIL を提出 • WebOnt Working Group　の立ち上げ ⇒ OWL のドラフト

  5. RDF（S)とOWLの関連 RDF(S) DAML+OIL ≒OWL • class-def • subclass-of • slot-def • subslot-of • domain • range • class-expressions • AND, OR, NOT • slot-constraints • has-value, value-type • cardinality • slot-properties • trans, symm

  6. Ontologies by Keyword Keyword URI academic department http://www.cs.umd.edu/projects/plus/DAML/onts/cs1.0.daml academic department http://www.cs.umd.edu/projects/plus/DAML/onts/cs1.1.daml Academic Positions http://www.daml.ri.cmu.edu/ont/homework/cmu-ri-employmenttypes-ont.daml access control primitives http://www.w3.org/2000/10/swap/pim/doc.rdf acronym http://orlando.drc.com/daml/Ontology/Thesaurus/CALL/current/ activity http://www.kestrel.edu/DAML/2000/12/OPERATION.daml Actors http://opencyc.sourceforge.net/daml/cyc.daml Actors http://www.cyc.com/2002/04/08/cyc.daml Actors http://www.cyc.com/cyc-2-1/cyc-vocab.daml address book http://www.w3.org/2000/10/swap/pim/contact.rdf agenda http://www.daml.org/2001/10/agenda/agenda-ont DAML Ontology Library (Ontology's by Keyword) http://www.daml.org/ontologies/keyword.html

  7. Baseball Ontology <rdf:RDF xmlns="http://www.daml.org/2001/08/baseball/baseball-ont#" xmlns:daml="http://www.daml.org/2001/03/daml+oil#" xmlns:oiled= "http://img.cs.man.ac.uk/oil/oiled#" xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#" xmlns:rdfs="http://www.w3.org/2000/01/rdf-schema#"> <daml:Ontology rdf:about=""> <rdfs:comment>baseball</rdfs:comment> <rdfs:comment>baseball ontology</rdfs:comment> <daml:versionInfo>"1.0"</daml:versionInfo> </daml:Ontology> <rdfs:Class rdf:about="http://www.daml.org/2001/08/baseball/baseball-ont#PostSeasonGame"> <rdfs:label>PostSeasonGame</rdfs:label> <rdfs:comment></rdfs:comment> <oiled:creationDate>12:55:22 26.08.2001</oiled:creationDate> <rdfs:subClassOf> <rdfs:Class rdf:about="http://www.daml.org/2001/08/baseball/baseball-ont#Game"></rdfs:Class> </rdfs:subClassOf> </rdfs:Class> <rdfs:Class rdf:about="http://www.daml.org/2001/08/baseball/baseball-ont#Inning"> <rdfs:label>Inning</rdfs:label> <rdfs:comment></rdfs:comment> <oiled:creationDate>12:22:49 26.08.2001</oiled:creationDate> <rdfs:subClassOf> <rdfs:Class rdf:about="http://www.daml.org/2001/08/baseball/baseball-ont#AggregateEvent"></rdfs:Class> </rdfs:subClassOf> <rdfs:subClassOf> <daml:Restriction> <daml:onProperty rdf:resource="http://www.daml.org/2001/08/baseball/baseball-ont#number"> </daml:onProperty> <daml:toClass> <rdfs:Class rdf:about="#http://www.w3.org/TR/xmlschema-2/#string"></rdfs:Class> </daml:toClass> </daml:Restriction> </rdfs:subClassOf> </rdfs:Class> <rdfs:Class rdf:about="http://www.daml.org/2001/08/baseball/baseball-ont#League"> <rdfs:label>League</rdfs:label> <rdfs:comment></rdfs:comment> <oiled:creationDate>12:35:38 26.08.2001</oiled:creationDate> <rdfs:subClassOf> <daml:Restriction> <daml:onProperty rdf:resource="http://www.daml.org/2001/08/baseball/baseball-ont#division"></daml:onProperty> <daml:toClass> <rdfs:Class rdf:about="http://www.daml.org/2001/08/baseball/baseball-ont#Division"></rdfs:Class> </daml:toClass> </daml:Restriction> </rdfs:subClassOf> </rdfs:Class> <rdfs:Class rdf:about="http://www.daml.org/2001/08/baseball/baseball-ont#Person"> <rdfs:label>Person</rdfs:label> <rdfs:comment></rdfs:comment> <oiled:creationDate>13:00:13 26.08.2001</oiled:creationDate> </rdfs:Class> <rdfs:Class rdf:about="http://www.daml.org/2001/08/baseball/baseball-ont#Player"> <rdfs:label>Player</rdfs:label> <rdfs:comment></rdfs:comment> <oiled:creationDate>12:25:42 26.08.2001</oiled:creationDate> <rdfs:subClassOf> <rdfs:Class rdf:about="http://www.daml.org/2001/08/baseball/baseball-ont#Employee"></rdfs:Class> </rdfs:subClassOf> </rdfs:Class> <rdfs:Class rdf:about="http://www.daml.org/2001/08/baseball/baseball-ont#Relief"> <rdfs:label>Relief</rdfs:label> <rdfs:comment></rdfs:comment> <oiled:creationDate>18:31:03 28.08.2001</oiled:creationDate> <rdfs:subClassOf> <rdfs:Class rdf:about="http://www.daml.org/2001/08/baseball/baseball-ont#LineupEvent"></rdfs:Class> </rdfs:subClassOf> </rdfs:Class> <rdfs:Class rdf:about="http://www.daml.org/2001/08/baseball/baseball-ont#Season"> <rdfs:label>Season</rdfs:label> <rdfs:comment></rdfs:comment> <oiled:creationDate>12:13:04 26.08.2001</oiled:creationDate> <rdfs:subClassOf> <daml:Restriction> <daml:onProperty rdf:resource="http://www.daml.org/2001/08/baseball/baseball-ont#year"></daml:onProperty> <daml:toClass> <rdfs:Class rdf:about="#http://www.w3.org/TR/xmlschema-2/#string"></rdfs:Class> </daml:toClass> </daml:Restriction> </rdfs:subClassOf> </rdfs:Class> <rdfs:Class rdf:about="http://www.daml.org/2001/08/baseball/baseball-ont#Series"> <rdfs:label>Series</rdfs:label> <rdfs:comment></rdfs:comment> <oiled:creationDate>12:56:25 26.08.2001</oiled:creationDate> <rdfs:subClassOf> <daml:Restriction> <daml:onProperty rdf:resource="http://www.daml.org/2001/08/baseball/baseball-ont#home"></daml:onProperty> <daml:toClass rdf:resource="http://www.daml.org/2001/08/baseball/baseball-ont#Team"/> </daml:Restriction> </rdfs:subClassOf> <rdfs:subClassOf> <daml:Restriction> <daml:onProperty rdf:resource="http://www.daml.org/2001/08/baseball/baseball-ont#away"></daml:onProperty> <daml:toClass

  8. ソフトウェア開発プロセスのためのオントロジー開発(人手）経験ソフトウェア開発プロセスのためのオントロジー開発(人手）経験

  9. Approach Issue First, Real software processes change dynamically... develop software process ontologies through the domain analysis using two case studies very difficult to maintain Afterwards, need the facility to correspond to changing process dynamically design the system which generate software processes interactively with a user Goal A Process-Centered Software Engineering Environment Using Ontologies

  10. C Invent output A+B Join Process A input α A α Append + reference Extract A’ B’ or B Development of Software Process Ontologies Step1. Extract real software processes and objects from the two cases Step2. Organize extracted processes and objects into distinct groups such that there are more similarity in meaning, while defining the software process scheme Step3. Give detailed definitions to each processes using software process scheme

  11. Process Ontology Invent with references Invent Invent without references Join with references Join Join without references activity Evaluate Append Review Examine Extract with references Extract Extract without references Focus on input,output, and reference roles Focus on reference, tool and agent roles

  12. Software Process Ontologies(conceptual hierarchy) Process Ontology including about 250 concepts Object Ontology including about 400 concepts

  13. Software Process Automation using ontologies ? ? Goal Object Start Object ? ? ? Retrieve processes satisfied with constraints Process Ontology Object Ontology

  14. Case Study A query about SＥP What SEP between the following input and output ? Input : Requirement Specification from S/W Subsystems Output : Detailed Design of Each Component of S/W Subsystems

  15. Detailed Design of Each Components of S/W Subsystems Design Basic Components of S/W Subsystems Requirement Specification from S/W Subsystems Design Detailed Components of I/F Subsystems Review All Basic Designs Together Design Detailed Components of S/W Subsystems Design Physical Aspects of DB Specify Testing S/W Integration Design Basic Components of I/F Subsystems Specify Testing Units of S/W Evaluate All Basic Designs Review All Detailed Designs Together Design Logical Aspects of DB Evaluate All Detailed Designs Initial Software Process Plan Candidates Output Input

  16. Detailed Design of Each Components of S/W Subsystems Design Basic Components of S/W Subsystems Requirement Specification from S/W Subsystems Design Detailed Components of I/F Subsystems Design Detailed Components of S/W Subsystems Design Physical Aspects of DB Specify Testing S/W Integration Design Basic Components of I/F Subsystems Specify Testing Units of S/W Evaluate All Basic Designs Review All Detailed Designs Together Design Logical Aspects of DB Evaluate All Detailed Designs Interim Software Process Plan Candidates Output Input

  17. Detailed Design of Each Components of S/W Subsystems Design Basic Components of S/W Subsystems Requirement Specification from S/W Subsystems Design Detailed Components of I/F Subsystems Design Detailed Components of S/W Subsystems Design Physical Aspects of DB Specify Testing S/W Integration Design Basic Components of I/F Subsystems Specify Testing Units of S/W Evaluate All Basic Designs Review All Detailed Designs Together Design Logical Aspects of DB Evaluate All Detailed Designs Final Software Process Plan Candidates Output Input

  18. Evaluation Applicability It turned out that the environment generated SPP candidates good for the user’s query with user interaction Issues and Future work • The environment can’t allocate human and time resource • The environment has not the facility that supports the user to add user’s specific processes • The methodology for building ontologies has no theoretical aspects,such as soundness

  19. 計算機可読型辞書とテキストコーパスからのオントロジー開発(半自動）経験計算機可読型辞書とテキストコーパスからのオントロジー開発(半自動）経験

  20. How to Build up Domain Ontologies with Less Cost • Taking Existing Information Resources DODDLE Project (a Domain Ontology rapiD DeveLopment Environment) exploiting a MRD Constructing up just a conceptual hierarchy

  21. Why not taking MRD to build up domain ontologies ? • Good News:MRD have many concepts. • Bad News:The concepts have been defined from the point of natural language processing (common sense) and so there are some concept drift between MRD and specific domain ontologies.

  22. Concept Drift DomainＢ Domain Ａ Reusable Part No reusable part because of concept drift MRD Domain

  23. DODDLE Spell-Match Spell-Matched Results A Set of Input Terms Selection of Best-match WordNet Initial Model Trimming Trimmed Model Domain Expert • Matched-Results Analysis • Trimmed- Results Analysis Extension of Modification Concept Hierarchy

  24. Root Best-Match Move Internal Node Stay Move Move Matched-Results Analysis

  25. A User re-constructs the sub-tree. A A Trimmed part B C 0 3 0 B C D B C Trimmed Model D D Initial Model Trimmed-Results Analysis

  26. Acquiring not only Conceptual Hierarchy but Concept Definitions DODDLE II Project (Extending DODDLE) Info. Resource:Domain-Specific Texts Technique: Co-occurence information WordSpace

  27. DODDLE-IIOverview a Set of Domain Terms MRD Domain Specific Texts Taxonomic Relationship Acquisition Module non-Taxonomic Relationship Learning Module WordNet Taxonomic Relationships non-Taxonomic Relationships Concept Specification Template extension & modification Domain Ontology

  28. Extracting Concept Relationshipsfrom Domain-Specific Texts WordSpace（Marti A. Hearst, Hinrich Schutze） • Words and phrases in texts can be expressed by vector representation containing co-occurrence statistics • Inner products among the vectors work as the similarity between the words and phrases. Context Similarity between concepts C1 and C2 … wi… wj…C1… wk… high value with similar words coming up around the concepts … wi… wj…C2… wk…

  29. Constructing WordSpace C C Context vector w … .. f8 f4 f3 f7 f8 f4 f1 f3 f4w f2 f5 f1 f7 f1 f5 .. … Word Vector W Vector representation of a concept 1. Extract Word 4-grams with high-frequency 2. Collocation Matrix 3. Context Vectors 4. Word Vectors ( WordSpace is a set of word vectors ) 5. Vector representations of all concepts Texts (4-gram array) Texts (4-gram array) Texts (4-gram vector array) … .. f8 f4 f3 f7 f8 f4 f1 f3 f8 f9 f2 f5 f1 f7 f1 f5 .. … … .. f8 f4 f3 f7 f8 f4 f1 f3 f4 f9 f2 f5 f1 f7 f1 f5 .. … w1 w2 w3 w4 w WordNet synset w collocation scope context scope f2 … .. f1 fn Collocation Matrix f1 A sum of 4-gram vectors around w .. Vector representation for each 4-gram f4 2 A sum of w in Texts … fn A sum of W in same synset 4-gram vector

  30. ConstructingConcept Specification Template a Set of Concept Pairs from non-TR Learning Module Ci Ca Ci Ci Ci Cc Cd Cb Taxonomic Relationships from TR Acquisition Module Cb Ci Ck ancestor, descendant and sibling Concept Specification Template Ci

  31. DODDLE-IIalphaon Perl/Tk(UNIX platform)

  32. A Case Study The Target Domain Contracts for the International Sale of Goods(CISG) Input Terms to TR module 46 Legal Terms from CISG Part-II Domain-Specific Texts to NonTR module full text ofCISG (about 10,000 words)

  33. 377Terms 113Terms Select Best Matches 46Terms Trimming 56Terms 61Terms Legal Ontology Initial Model Trimmed Model The paths from Spell Matched Nodes to the Root of WordNet Results Modification

  34. Taxonomic Relationships for input termsconstructed from TR Acquisition Module

  35. Support Ratio Support ratio: How match is included in final domain ontology the intermediate products at each DODDLE activity.

  36. ExtractingConcept Relationships setting value for WordSpace the concept pairs extracted according to context similarity (threshold 0.9993)

  37. Domain Experts Modifying Concept Specification Templates ex) non-Taxonomic Relationships for “assent” non-Taxonomic Relationships from the template Concept Specification Template assent assent AGENT : person LEGAL-SEQUENCE : offer ANTONYM : withdrawal DE Modification

  38. Recall & Precision of Concept Specification Templates recall precision The number of extracted concept pairs

  39. Conclusions DODDLE II: A Domain Ontology Construction Support Environment using MRD and Domain-Specific Texts Legal experts appreciate DODDLE II to some extent. Future Work • How to Decide Threshold for CS • How to Identify NTR • Another DM method instead of WordSpace • Another Case Studies with Large Scale

  40. OWLとオントロジー開発 • OWLとオントロジー開発⇔ UMLとソフトウェア開発 • OWL＋RDF(S)＋RDF⇒中味のあるオントロジー • オントロジー構築支援ツールが必要 • オントロジーの重量⇔スケーラビリティ

More Related