The WebKB set of tools

Knowledge Modelling and Retrieval on the WWW

Philippe A. MARTIN

(e-mail: pm .@. phmartin dot info)

Griffith University, School of Information Technology

Work supported by the Defence Science And Technology Organisation (DSTO)

Talk Structure

• Purpose: solving some IR and KA problems
  
  
• A solution: combining generic facilities of IR and KA
  
  
• Architecture
  
  
• Limitations
  
  
• Demonstration
  
  
• Conclusion

Some problems of most current IR tools

The semantic content of documents is not represented:

• No question answering, just document retrieval
  
  
• No document generation/synthesis

Some problems of most current KA tools

KA: lots of searches, representations and (re)organisations
in documents and in the KB, but

• poor systems for knowledge retrieval/generation
  
  
• poor systems for document element retrieval/annotation
  
  

Some problems of cooperation tools

• Threaded discussion servers (e.g. Hypernews): no KR
  
  
• Document annotation servers (e.g. CoNote): no KR
  
  
• Shared document servers (e.g. Futplex): no KR
  
  
• SOAP servers/clients (e.g. with PICS): no KR, no annotation
  
  
• Knowledge-based WWW-traders (e.g. AlephWeb
  NetRepository, AI-trader): document retrieval, no annotation

Architecture

• WWW-accessible knowledge processor (+ warehouse)
  
  
• Predefined ontologies (in HTML documents / warehouse)
  
  
• Graphical interfaces: Netscape + Javascript
  
  

Interfaces

• Interfaces for Web-document/knowledge handling commands
  
  
• Interface for building indexations of DES by knowledge
  
  
• Conceptual graph textual/graphic editors
  
  
• Hierarchy browser

Languages

• Combinable document/knowledge handling/retrieval commands
  
  
• Commands or representations are embeddable in documents
  
  
• Any WWW document element may be indexed
  
  
• Various notations: CGs, frame-oriented CGs, HTML structures
  
  
• Types may or may not be declared

Ontologies

• 200 top-level concept types -> consistency, reuse
  
  
• 200 basic relation types -> direct use, consistency
  
  
• 90,000 NL concept types -> direct use, retrieval, reuse

Limitations

• For large-scale IR: no automatic indexation of documents
  
  
• For KA: no rule system
  
  
• For CSCW: no real-time interaction between users

Conclusion

Web-accessible complementary and combinable facilities.

http://www.webkb.org/
http://www.int.gu.edu.au/~phmartin/WebKB/