Extracting Semantic Constraint from Description Text for Semantic Web Service Discovery

1
Extracting Semantic Constraint from Description
Text for Semantic Web Service Discovery
Dengping Wei, Ting Wang, Ji Wang, and Yaodong
Chen Reporter Ting Wang Department of Computer
Science and Technology School of
Computer National University of Defense
Technology, China tingwang_at_nudt.edu.cn
2
Outline

• Motivation
• Semantic Constraint for SWS Discovery
• Extracting Semantic Constraint
• Matching Algorithm
• Experiment Results
• Conclusions and Future Work

3
Motivation

• Various Semantic Web Service (SWS) description
  ontologies or frameworks
• e.g. OWL-S, WSMO, WSDL-S, SAWSDL.
• Various SWS matchmakers
• logic based semantic IOPE matching
• inputs(I), outputs(O), preconditions/assumptions(P
  ) and effects/postconditions(E)
• logic based semantic Input/Output matching

4
Motivation

• Most current SWS matchmakers treat the SWS
  signature as a set of concepts
• not sufficient to discover SWS
• two services with similar semantics may fail to
  match
• two services with the same input and output
  concepts may have essential differences in
  semantics
• which may not be detected by logic based
  reasoning.

5
Motivation

• Many recent researches have explored various
  information to complement service I/O concepts
  for SWS matchmaking
• The ranked matching algorithm Jaeger, et al.
  2005
• A hybrid method for SWS discovery Klusch, et al.
  2006
• SWS matchmaking based on iSPARQL Kiefer, et al.
  2008
• Hull, et al. 2006 describes the relationships
  and uses OWL ontologies

6
Motivation

• The relationships between the service I/O
  concepts can be helpful for expressing the
  semantics of services.

7
Motivation

• Our idea
• add some restriction relationships to the
  interface concepts
• to enhance the semantic description of services.
• extract restriction relationships
• those relationships not defined in the domain
  ontology.
• from the service description text
• automatically
• perform the matching on the service I/O concepts
  and their semantic constraints represented by a
  constraint graph

8
Outline

• Motivation
• Semantic Constraint for SWS Discovery
• Extracting Semantic Constraint
• Matching Algorithm
• Experiment Results
• Conclusions and Future Work

9
Semantic Constraint for SWS Discovery

• Observation
• the domain of concept is not specified
• the price of a book
• the price of a flight ticket
• the property of concept is not specified
• the food with the maximum price
• the food with brand Coca Cola
• the relationship between concepts is not
  specified
• the food contained in a certain grocery store
• the food sold by a certain grocery store

10
Semantic Constraint for SWS Discovery

• The semantics of SWS will be better clarified
• if the constraint relationships of the concepts
  have been annotated

11
Semantic Constraint for SWS Discovery

• Definition of a statement ltSC,CT,OCgt
• SC (Subject Concept)
• subject of the statement
• usually corresponds to the service I/O concepts.
• OC (Object Concept)
• object of statement
• described as another concept or a literal.
• CT (Constraint Type)
• predicate of the statement
• identifies the property or characteristic of the
  subject concept that the statement specifies.

12
Constraint Types Definition

• CT (Constraint Type)
• three important abstract constraint types
• isPropertyObjectOf Constraint
• triple ltA, isPropertyObjectOf,Bgt means that
  concept A is a property object of concept B.
• hasPropertyObject Constraint
• this constraint relation is the inverse of
  isPropertyObjectOf.
• Operation Constraint
• triple ltA, Operation, Bgt means that two concepts
  entities have a certain association between them
• lt Book, published by, Springer gt
• the books that are published by Springer

13
Constraint Graph Definition

• Definition
• Let C be a set of concepts, a directed constraint
  graph can be described as ConstraintGraph(C)
  ltSC,CT,OCgtSC ? C.

14
Outline

• Motivation
• Semantic Constraint for SWS Discovery
• Extracting Semantic Constraint
• Matching Algorithm
• Experiment Results
• Conclusions and Future Work

15
Extracting Semantic Constraint
Stanford PCFG Parser
Fig. 2. Semantic constraint extraction
16
Extracting Semantic Constraint

• Candidate Constituent Detection
• Constraint Constituents Filtering
• Extracting Modifier

17
Extracting Semantic Constraint

• Candidate Constituent Detection
• observation
• the constraints of a key-word are probably
  contained in the phrase whose head word is the
  keyword.
• detect all such phrases by propagating the
  key-word from the bottom to the top of the
  syntactic tree.
• the propagation path is expressed as a sequence
  of interior nodes in the parsing tree
• e.g. a node sequence NP NP in the example is
  the propagation path of key-word price.

18
Constraint Constituents Filtering and Extracting
Modifier
19
Outline

• Motivation
• Semantic Constraint for SWS Discovery
• Extracting Semantic Constraint
• Matching Algorithm
• Experiment Results
• Conclusions and Future Work

20
Matching Algorithm

• Constraint Graph Matching(CGM)
• where P is the number of triples in
  ConstraintGraph(Cr )
• P the number of triples in ConstraintGraph(Cs)
• function TripleMatch(RTi, STi) to estimate the
  match between two triples RTi and STj.

21
Matching Algorithm

• Triples Matching
• two triples are matched and the degree of match
  can be measured
• if all the three elements in each triple are
  relative

22
Matching Algorithm

• Concept Matching
• matching five different levels
• Exact match r s.
• Plug-in match r ?Ascendant (s) ? s?Descendant(r)
• Subsumed-by match s?Ascendant(r)?r?Descendant(s)
• Intersect match
• Fails

23
Outline

• Motivation
• Semantic Constraint for SWS Discovery
• Extracting Semantic Constraint
• Matching Algorithm
• Experiment Results
• Conclusions and Future Work

24
Experiment Results

• OWL-S TC v2
• 576 web services from 7 domains
• 28 queries with their relevance sets.
• http//www-ags.dfki.uni-sb.de/klusch/owls-mx/
• Two sets of web services
• dataset1
• the semantic constraints of the output concepts
  in request and web service are manually annotated
  by two people
• mainly described by service I/O concepts
• dataset2
• the semantic constraints of concepts are
  automatically extracted using the method
  represented above

25
Experiment Results

• Klusch et al. 2006
• OWLS-M0 is a pure logic based matchmaker on the
  service I/O concepts
• OWLS- M4 is reported to be the best-performing
  matchmaker variant of the OWLS-MX matchmaker
• M0InOutConstraint matchmaker uses CGM to filter
  the results of OWLS-M0 on Dataset1
• M0AutoConstraint matchmaker uses CGM to filter
  the results of OWLS-M0 on Dataset2
• M4InOutConstraint matchmaker uses CGM to filter
  the results of OWLS-M4 on Dataset1
• M4AutoConstraint matchmaker uses CGM to filter
  the results of OWLS-M4 on Dataset2

26
Experiment Results
InOutConstraint
OWLS-M4
OWLS-M0
The performance on Dataset1
27
Experiment Results
M4InOutConstraint
M0InOutConstraint
OWLS-M4
OWLS-M0
The performance on Dataset1
28
Experiment Results
M4AutoConstraint
M0AutoConstraint
OWLS-M4
AutoConstraint
OWLS-M0
The performance on Dataset2
29
Outline

• Motivation
• Semantic Constraint for SWS Discovery
• Extracting Semantic Constraint
• Matching Algorithm
• Experiment Results
• Conclusions and Future Work

30
Conclusion

• Introduce semantic constraints for service I/O
  concepts
• enhancing the semantics of web service
• Extract semantic constraints automatically from
  the parsing trees of the description text
• Use constraint graph to describe the semantic
  constraints of the service I/O concepts
• A matching algorithm for the constraint graph

31
Future Work

• Finding more effective extraction method
• to get better results of extraction
• Extract more constraint relationships for the
  concepts
• web service can be represented by a more
  complicated graph
• more sophisticate matching algorithm

32

• Thank you!