DYKOR - A METHOD FOR GENERATING THE CONTENT OF EXPLANATIONS IN KNOWLEDGE SYSTEMS

The paper presents a methodology for improving the organization of kno wledge bases and demonstrates its application for generating the conte nt of explanations. The DyKOr (Dynamic Knowledge Organization) method combines information that is usually available through execution trace s with existing domain knowledge using techniques from machine learnin g including knowledge compilation, explanation-based learning, and con ceptual clustering. These techniques allow the separation of the knowl edge needed to solve a problem from that which is not required, and th e identification of information that is related to the problem but is not explicitly stated. Thus, the analysis performed through the method ology can considerably improve the quality and content of explanations . The paper describes the implementation of the methodology and how it can be integrated into typical rule-based expert systems. Illustratio ns of how the method can be used to produce the content for explanatio ns are presented in the context of typical consultation and problem so lving expert systems. A discussion of how the information produced by the method can be used to prepare explanations for users with differen t levels of expertise is also presented.