Representing Problem-Solving for Knowledge Refinement

Abstract

Knowledge refinement tools seek to correct faulty
\nknowledge based systems (KBSs) by identifying
\nand repairing potentially faulty rules. The goal
\nof the KrustWorks project is to provide a source
\nof refinement components from which specialised
\nrefinement tools tailored to the needs of a range
\nof KBSs are built. A core refinement algorithm
\nreasons about the knowledge that has been applied, but this approach demands general knowledge structures to represent the reasoning of a
\nparticular problem solving episode. This paper investigates some complex forms of rule interaction
\nand defines a knowledge structure encompassing
\nthese. The approach has been applied to KBSs
\nbuilt in four shells and is demonstrated on a small
\nexample that incorporates some of the complexity
\nfound in real applications.