Compositiondriven Filtering and Selection of Semantic Web Services

1
Composition-driven Filtering and Selection of
Semantic Web Services

• Evren Sirin, Bijan Parsia, Jim Hendler
• Mindswap Research Group
• University of Maryland, College Park

2
Introduction

• There are web services around!
• Lets compose them together
• A user oriented tool
• Interactive, intuitive, easy to use
• Use the current standards
• Off-the-shelf technologies
• Handle devices

3
Semantic Descriptions

• Rich service descriptions
• Locating services based on capability
• OWL-S descriptions
• We started with DAML-S 0.6!
• Generate from WSDL
• AtomicProcess - WSDL Operation
• Profile - Text descriptions from WSDL
• Grounding - Easy to do!

4
WSDL to OWL-S
5
Grounding Issues

• Parameter types
• WSDL XML Schema
• OWL-S OWL Classes

ontstreet
ontstate
RDF
AddressProvider
MapQuest Service
OWL-S


tate

AddressProvider
MapQuest Service
WSDL
SOAP
6
Marshalling/Unmarshalling

• Use XSLT stylesheet
• Doesnt need separate processor
• Provides extensive functions
• Solves Decker problem in a way
• Some problems
• Various RDF/XML serializations
• Requires deep understanding of both structures
• Data conversion overhead

Contributions by Ryusuke Masuoka from Fujitsu
Labs of America, College Park
7
Other alternatives

• Enrich WSDL descriptions
• Include RDF specification
• Incorporate RDF query in XSLT
• RDQL or Versa
• Higher level mapping language
• Mapping Definition Language (MDL)
• Can be compiled to XSLT

8
Interactive Composition Approach

• Semi-automatic process
• Goal oriented
• Guide a user through steps of building a
  composition
  
• Start with a service
• Present matching services
• Filter possibilities

9
Web Service Composer
10
Matching By IOPEs

• Use functional attributes
• Input/Output/Precondition/Effect (IOPEs)
• Reason about parameter types
• Use Description Logic (DL) subsumption
• Exact matches
• Generic matches
• Also referred as PlugIn
• Relaxed matches
• Exception handling

11
Selection by Filtering

• Profile hierarchies
• Class-hierarchical classification
• Provides a taxonomy for services
• Use non functional attributes
• Defined by ServiceParameters
• Expandable list of properties
• Attributes linked to concepts in ontologies

12
Filtering
13
Filtering Forms
14
Handling Devices

• Use any network accessible program
• Devices providing their functionality as
  services
  
• Sensor, printer, projector, etc.
• Pervasive computing
• Use OWL-S to describe services
• Use OWL to markup device attributes
• Can be used for filtering

15
Saving Composition
16
Different Ontologies

• Services referring different ontologies
• No link between ontologies
• Define (partial) mappings
• Semantic Mappings (OWL constructs)
• More generic axioms (OWL Rules)
• Arbitrary computation
• Transformation between types
• Doesnt require full consistency between
  ontologies
  

17
Ontology Translation Services

• Wrap transformations as services
• Easy integration to the system
• Special profiles for translators
• Hide from user
• Create composed services automatically

18
Linking Ontologies - OntoLink
19
Future directions

• Expressive service descriptions
• Need to go beyond input/output
• Too much responsibility on user
• Domain knowledge requirement
• Planner help user make decisions
• HTN Planning
• Learning from past activity
• History of compositions
• Complex workflows
• Conditionals, iterations, etc.