Empirical Analysis and Refinement of Expert System Knowledge Bases

Knowledge base refinement is the modification of an existing expert system knowledge base with the goals of localizing specific weaknesses in a knowledge base and improving an expert systems performance. Systems that automate some aspects of knowledge base refinement can have a significant impact on the related problems of knowledge base acquisition, maintenance, verification, and learning from experience. The SEEK system was the first expert system framework to integrate large-scale performance information into all phases of knowledge base development and to provide automatic information about the refinement. A recently developed successor system, SEEK2, significantly expands the scope of the original system in terms of generality and automated capabilities. Based on promising results using the SEEK approach, we believe that significant progress be made in expert system techniques for knowledge acquisition, knowledge base refinement, maintenance, and verification. We are proposing to demonstrate a rule refinement system in an application of the diagnosis of complex equipment failure. The expected candidate application is computer network troubleshooting. The expert system should demonstrate the following advanced capabilities 1 automatic localization of knowledge base weaknesses 2 automatic repair refinement of poorly performing rules 3 automatic verification of new knowledge base rules and 4 some automatic learning capabilities.