OWLbased Semantic Conflicts Detection and Resolution for Data Interoperability

1
OWL-based Semantic Conflicts Detection
and Resolution for Data Interoperability

• Changqing Li, Tok Wang Ling
• Department of Computer Science
• School of Computing
• National University of Singapore

2
Outline

• Introduction
• Preliminary and motivation
• OWL-based Semantic Conflicts Detection and
  Resolution
• Conclusion
• Q A

3
Introduction

• Data interoperability and integration is a
  long-standing challenge to the database research
  community.
• Ontology provides sharing knowledge among
  different data sources
• Clarify the semantics of information.
• Provide a way to solve the interoperability
  problem in database integration

4
Introduction (Cont.)

• OWL is being promoted as a standard for web
  ontology language
• In the future a considerable number of ontologies
  will be created based on OWL.
• Therefore automatically detecting semantic
  conflicts based on OWL will greatly expedite the
  step to achieve semantic interoperability, and
  will greatly reduce the manual work to detect
  semantic conflicts.

5
Ontology Definition

• An ontology defines the basic terms and relations
  comprising the vocabulary of a topic area, as
  well as the rules for combining terms and
  relations to define extensions to the vocabulary
  1.

1. Robert Neches, Richard Fikes, Timothy W.
Finin, Thomas R. Gruber, Ramesh Patil, Ted E.
Senator, William R. Swartout Enabling Technology
for Knowledge Sharing. AI Magazine 12(3) pp36-56
(1991)
6
Ontology Language

• SHOE
• RDF
• RDFS
• DAMLOIL
• OWL

7
SHOE

• The Simple HTML Ontological Extensions (SHOE) 2
  extends HTML with machine-readable knowledge
  annotated.

2. Sean Luke and Jeff Heflin SHOE
Specification 1.01. http//www.cs.umd.edu/projects
/plus/SHOE/spec.html
8
RDF

• Resource Description Framework (RDF) 3 is a
  recommendation of W3C for Semantic Web 4
• It defines a simple model to describe
  relationships among resources in terms of
  properties and values.
• SVO form (Subject-Verb-Object)
• Resource-property-Value

• 3. Ora Lassila and Ralph R. Swick Resource
  description framework (RDF).
• http//www.w3c.org/TR/WD-rdf-syntax
• 4. The SemanticWeb Homepage.
  http//www.semanticweb.org

9
RDF (Cont.)
10
RDFS

• RDF Schema (RDFS) 5, the primitive description
  language of RDF
• Provide some basic primitives
• subClassOf
• subPropertyOf

5. Dan Brickley and R.V. Guha. Resource
Description Framework (RDF) Schema Specification
1.0, W3C Candidate Recommendation 27 March 2000.
http//www.w3.org/TR/rdf-schema/
11
DAMLOIL

• DARPA Agent Markup Language (DAML) 6
• To facilitate the semantic concepts and
  relationships understood by machines
• Ontology Inference Layer (OIL) 7
• Extends RDFS with additional language primitives
  not yet presented in RDFS.
• DAMLOIL 8 are the successors of RDFS
• Combination of DAML and OIL
• More semantic rich primitives are defined

• 6. The DARPA Agent Markup Language Homepage.
• http//daml.semanticweb.org/
• 7. The Ontology Inference Layer OIL Homepage.
• http//www.ontoknowledge.org/oil/TR/oil.long.html
  
• 8. DAMLOIL Definition. http//www.daml.org/2001/0
  3/damloil

12
OWL

• DAMLOIL is evolving as OWL (Web Ontology
  Language) 9.
• OWL is almost the same as DAMLOIL
• Some primitives of DAMLOIL are renamed in OWL
  for easier understanding.
• e.g., sameClassAs is changed to
  equivalentClass

9. Frank van Harmelen, Jim Hendler, Ian
Horrocks, Deborah L. McGuinness, Peter F.
Patel-Schneider and Lynn Andrea Stein. OWL Web
Ontology Language Reference. http//www.w3.org/TR/
owl-ref/
13
Primitives of OWL

• owl before is the namespace
• owlequivalentClass
• owleuqivalentProperty
• owlsameIndividualAs
• owldisjointWith
• owldifferentFrom

14
Our Extension of OWL (EOWL)

• We extend OWL with the following primitives
• eowlorderingProperty
• eowloverlap
• eowlproperSubClassOf
• eowlproperSubPropertyOf

15
OWL-based Semantic Conflicts Cases

• A. Name conflicts
• B. Order sensitive conflicts
• C. Scaling conflicts
• D. Whole and part conflicts
• E. Partial similarity conflicts
• F. Swap conflicts

16
A. Name conflicts

• Example A. two distributed data warehouses
• one is used to analyze the United States market
• country, state, city and district
• and the other is used to analyze the China market
• country, province, city and county
• Based on the context
• provicnce is defined equivalent to State
  using the OWL primitive owlequivalentClass.
• To resolve this conflict, one name needs to be
  changed. Change to the referenced name.

17
A. Name conflicts (Cont.)
ltowlClass rdfID"Province"gt
ltrdfslabelgtProvincelt/rdfslabelgt
ltowlequivalentClass rdfresource"State"/gt lt/owl
Classgt
Fig. A. Detection of synonym conflicts

• owlequivalentClass is the indicator to detect
  synonym conflicts
• Change to State as which is referenced in the
  ontology definition.

18
A. Name conflicts (Cont.)

• Case A. Synonyms. The OWL primitives
  owlequivalentClass, owlequivalentProperty
  and owlsameInvidualAs are indicators to detect
  this case.
• Conflict Resolution Rule A. If synonym conflicts
  are detected, different attribute names with the
  same semantics need to be translated to the same
  name (referenced name) for smooth data
  interoperability.

19
B. Order sensitive conflicts

• Example B. Consider the highest three scores of a
  course.
• The highest three scores of course A are listed
  as 90, 95, 100 at ascending order,
• The highest three scores of course B are listed
  as 98, 95, 93 at descending order.
• The highestThreeScores is defined as an
  eowlorderingProperty in the ontology
• The sequences of the highest three scores for
  course A and B should be adjusted both to
  ascending order or descending order.
• Adjust to the sequence of the first one by
  default, e.g. the sequence of course A

20
B. Order sensitive conflicts (Cont.)
lteowlorderingProperty rdfID"highestThreeScores"
gt ltrdfslabelgthighest three scores of a
courselt/rdfslabelgt ltrdfsdomain
rdfresource"Course"/gt ltrdfsrange
rdfresource"xsdinteger"/gt lt/eowlorderingProper
tygt
Fig. B. Detection of order sensitive conflicts

• We can further define the ascendant or descendant
  order for more precise semantics.

21
B. Order sensitive conflicts (Cont.)

• Case B. Order sensitive. EOWL primitive
  eowlorderingProperty and RDF primitive
  rdfSeq are indicators to detect this case.
• Conflict Resolution Rule B. If order sensitive
  conflicts are detected, we need to adjust the
  member sequence according to the same criterion
  for smooth data interoperability, the sequence of
  the first one by default.

22
C. Scaling conflicts

• Example C. Consider two database schemas
• Product(ID, Price)
• Product(ID, Price)
• One price may refer to the US dollars, while the
  other may refer to the Singapore dollars. Figure
  4 shows some concepts about a currency ontology
  price is defined
• Translate the price to refer to the same currency
  unit. The unit of the first one by default.

23
C. Scaling conflicts (Cont.)
ltowlDatatypeProperty rdfID"price"gt
ltrdfsdomain rdfresource"Product"gt
ltrdfsrange rdfparseType"Resource"gt
ltrdfvalue/gt ltcurrencyCurrencyUnit/gt
lt/rdfsrangegt lt/owlDatatypePropertygt
Fig. C. Detection of scaling conflicts
24
C. Scaling conflicts (Cont.)

• Case C. Semantic conflicts may exist if the value
  of a data type property comprises both value and
  unit (Scaling). RDF primitive rdfparseType"Reso
  urce" and OWL primitive owlDatatypeProperty
  are indicators for this case.
• Conflict Resolution Rule C. If scaling conflicts
  are detected, the value should be translated to
  refer to the same unit for smooth data
  interoperability. The first unit by default.

25
D. Whole and part conflicts

• Example D. Consider schemas
• Person(ID, name)
• Person(ID, surname, givenName)
• surname and givenName are both defined as the
  proper sub property of name using
  eowlproperSubClassOf
• eowlproperSubClassOf has clearer semantics
  than rdfssubClassOf because rdfssubClassOf
  is ambiguous with two meanings
  eowlproperSubClassOfand owlequivalentClass.
  
• Divide the whole attribute name to the part
  attributes surname and givenName
• Or combine the part attributes surname and
  givenName together in the correct sequence to
  form the whole attribute name.

26
D. Whole and part conflicts (Cont.)
ltrdfProperty rdfID"surname"gt
lteowlproperSubPropertyOf rdfresource"name"gt lt/
rdfPropertygt
Fig. D1. Detection of whole and part conflicts
ltrdfProperty rdfIDgivenname"gt
lteowlproperSubPropertyOf rdfresource"name"gt lt/
rdfPropertygt
Fig. D2. Detection of whole and part conflicts
27
D. Whole and part conflicts (Cont.)

• Case D. Semantic conflicts may exist if one
  concept is completely contained in another
  concept (Whole and part). EOWL primitives
  eowlproperSubClassOf, eowlproperSubPropertyOf
  are indicators to detect this case.
• Conflict Resolution Rule D. If whole and part
  conflicts are detected, the whole attributes
  should be divided into part attributes or the
  part attributes should be combined together to
  whole attributes for smooth data interoperability.

28
E. Partial similarity conflicts

• Example E. integration ResearchAssistant and
  GraduateStudent
• The relationship between research assistant and
  graduate student is overlap because some research
  assistants are also graduate students,
• but not all research assistants are graduate
  students,
• and not all graduate students are research
  assistants.
• After integration, there should be three schemas
  
• Research Assistant but not Graduate Student
  RNotG
• Graduate Student but not Research Assistant
  GNotR
• both Research Assistant and Graduate Student
  RAndG

29
E. Partial similarity conflicts (Cont.)
ltowlClass rdfID"ResearchAssistant"gt
lteowloverlap rdfresource"GraduateStudent"/gt lt/
owlClassgt
Fig. E. Detection of partial similarity conflicts
30
E. Partial similarity conflicts (Cont.)

• Case E. Semantic conflicts may exist if two
  concepts are overlapped (Partial similarity).
  EOWL primitive eowloverlap is indicators to
  detect this case.
• Conflict Resolution Rule E. If partial similarity
  conflicts are detected, the overlap part should
  be separated before integration.

31
F. Swap conflicts

• Example F. Continued from Example A
• In China, county is contained in city (city has
  larger area)
• In US, city is contained in county (county has
  larger area).
• The domain (County) of property
  regioncontainedIn in the China ontology is
  just the range of the same property
  regioncontainedIn in the US ontology
• The range (City) of property regioncontainedIn
  in the China ontology is just the domain of the
  same property regioncontainedIn in the US
  ontology.
• We can add China. or US. before City and
  County for smooth data interoperability.

32
F. Swap conflicts (Cont.)
ltowlClass rdfID"County"gt
ltregioncontainedIn rdfresource"City/gt lt/owlC
lassgt
Fig. F1. Detection of swap conflicts (the
relationship between city and county in the China
ontology)
ltowlClass rdfID"City"gt ltregioncontainedIn
rdfresource"County/gt lt/owlClassgt
Fig. F2. Detection of swap conflicts (the
relationship between city and county in the US
ontology)
33
F. Swap conflicts (Cont.)

• Case F. Semantic conflicts may exist if the
  domain of a property in the first ontology is the
  range of the same property in the second
  ontology, and the range of the property in the
  first ontology is the domain of the same property
  in the second ontology (Swap).
• Conflict Resolution Rule F. If swap conflicts are
  detected, context restrictions (see Example F)
  should be added to the schema for smooth data
  interoperability.

34
Conclusion

• We extend OWL with several primitives which have
  clearer semantics
• summarize several cases based on OWL in which
  semantic conflicts are easily to be encountered
• The conflict resolution rules for each case are
  presented.
• In the future, OWL will be frequently used to
  build ontologies, and this paper provides a
  computer-aid approach to detect and resolve
  semantic conflicts for smooth data
  interoperability.

35
References

• 1. Robert Neches, Richard Fikes, Timothy W.
  Finin, Thomas R. Gruber, Ramesh Patil, Ted E.
  Senator, William R. Swartout Enabling Technology
  for Knowledge Sharing. AI Magazine 12(3) pp36-56
  (1991)
• 2. Sean Luke and Jeff Heflin SHOE
  Specification 1.01. http//www.cs.umd.edu/projects
  /plus/SHOE/spec.html
• 3. Ora Lassila and Ralph R. Swick Resource
  description framework (RDF).
• http//www.w3c.org/TR/WD-rdf-syntax
• 4. The SemanticWeb Homepage.
  http//www.semanticweb.org
• 5. Dan Brickley and R.V. Guha. Resource
  Description Framework (RDF) Schema Specification
  1.0, W3C Candidate Recommendation 27 March 2000.
  http//www.w3.org/TR/rdf-schema/
• 6. The DARPA Agent Markup Language Homepage.
• http//daml.semanticweb.org/
• 7. The Ontology Inference Layer OIL Homepage.
• http//www.ontoknowledge.org/oil/TR/oil.long.html
  
• 8. DAMLOIL Definition. http//www.daml.org/2001/0
  3/damloil
• 9. Frank van Harmelen, Jim Hendler, Ian
  Horrocks, Deborah L. McGuinness, Peter F.
  Patel-Schneider and Lynn Andrea Stein. OWL Web
  Ontology Language Reference. http//www.w3.org/TR/
  owl-ref/

36
Thank you