15th European Conference on Artificial Intelligence
|July 21-26 2002 Lyon France|
The pioneer systems for rule refinement are SEEK and SEEK2. Unlike TEIRESIAS, which also has been designed for the acquisition of new inference rules, the systems SEEK and SEEK2 are devoted to the refinement of rules for a rheumatology rule-base (a medical diagnosis application). This article investigates the general refinement completeness of SEEK/SEEK2. A rule refinement system is complete if it solves every possible refinement problem. SEEK2 has refinement heuristics for coping with generalization and specialization problems. Complete rule refinement systems should also have refinement capabilities for tackling a third refinement class to be called context refinement. On the syntactic level, the rheumatology rules which are subject of the SEEK/SEEK2 refinements have no logical negation in their if-parts. On the semantic, findings which are to be interpreted by the rheumatology rule-base are represented in positive form only. This seems to be the reason for the incompleteness of SEEK2 with regard to context problems, i.e., there was no need for context refinement heuristics. A complete rule refinement system must employ methods for contextualization, as well as generalization and specialization.
Keywords: Knowledge Acquisition, Satisfiability Testing, Intelligent User Interfaces, Meta-Heuristics for AI, Verification and Validation, Knowledge Discovery
Citation: Hans-Werner Kelbassa: Context Refinement - Investigating the Rule Refinement Completeness of SEEK/SEEK2. In F. van Harmelen (ed.): ECAI2002, Proceedings of the 15th European Conference on Artificial Intelligence, IOS Press, Amsterdam, 2002, pp.205-209.