A Methodology for Conducting Knowledge Discovery on the Semantic Web

1
A Methodology for Conducting Knowledge Discovery
on the Semantic Web

• Dimitrios Koutsomitropoulos, Markos Fragakis,
  Theodore Papatheodorou
• HPCLab, University of Patras

2
Introduction

• Why is KD important? (in the SW context)
• Intelligent search
• More efficient retrieval of resources
• Service discovery and matchmaking
• No clear and sound method yet exists
• Fragmented approaches
• Know-how but no how-to!
• Need for a well-identified procedure

3
Semantic Web Knowledge Discovery

• Definition The ability to discover information
  that is not explicitly expressed in Web Ontology
  documents
• Product of reasoning that leverages existing
  information to infer logical conclusions

4
Our ProposalA Methodology that aims to be

• Intuitive
• Allowing friendly query construction
• Withholding details of actual KB querying
• Declarative
• No prior knowledge of ontology contents
  structure required
• Expressive
• Allowing powerful inferences on ontology
  documents
• Knowledge Discovery Interface (KDI) as a
  prototype implementation

5
Previous Work

• OWLIR, TAP, Artequakt
• Not always performing SWKD
• Not clarifying or prescribing a procedure towards
  their KD approaches
• Query Languages (DQL, OWL-QL)
• Formal querying communication protocols
• Not declarative enough
• Not a way to conduct inference

6
The Methodology

• Three phases
• Selection of appropriate logical formalism
• Criteria for Web-environment suitability
• Selection of a specific inference engine

7
Formalism

• Description Logics
• OWL DL ? SHOIN(D) in polynomial time
• SHOIN (D) SHIN(D) (incomplete) Horrocks,
  Schneider 03
• FOL (Theorem provers)
• Hoolet, Surnia, JTP
• Manual insertion and verification of equivalent
  axioms
• Rule based reasoning systems
• OWLLisaKB, SNOBASE, Jena
• Up to OWL-Lite only

8
Special Requirements for the SW

• Taxonomic Method for KD
• DL and logic enabled DB literature and history
• Using only concepts and classification (TBox)
• Approximating instances with concepts
• Not enough for the Web!
• The Web Need for ABox reasoning
• Arbitrary detail of semantic structures ?
  Fine-grained analysis of instances
• Instance based method functions and components
• Instance checking
• Role fillers retrieval

9
Selecting a Reasoner
10
Putting all in action The KDI
11
Query Composition

• Browse/explore concept taxonomy
• Select a concept
• Select an instance and a role
• View/Retrieve results

Instances and roles adapt to concept selection
12
Component Architecture and Query Decomposition
13
Experimental Results

• A series of inferences through the KDI
• on the CIDOC-CRM
• Domain ontology for cultural heritage
• Ported to OWL
• Semantically extended and enriched with
  OWL-specific structures
• Cardinality restrictions, inverse roles,
  existential and universal quantifications
• Up to OWL-DL level
• Created concrete instances modeling cultural
  information (Piet Mondrian)

14
Example 1

• Using Value Restriction
• Successful!

15
Example 2

• Using Existential Quantification and Nominals
• Unsuccessful due to nominal approximation

16
Retrieval Capabilities

• Using Concrete Domains (strings, URLs)
• Retrieving actual resources

17
Conclusions

• SWKD
• Well-studied background
• Adaption to the Web mature and almost
  standardized
• Suggest and verify a sound-yet-flexible
  methodology
• Decisions at each stage may change as
  technologies evolve
• KDI as a prototype implementation
• Declarative and intuitive way for conducting
  inferences
• Problem not yet fully solved!
• Expressivity and scalability concerns
• Evolution of systems
• Recent and upcoming standards (WRL, SWRL, OWL)