Knowledge base clustering for KBS maintenance

Clustering the rules in a knowledge-based system (KBS) on the basis of their static distance (lexical information) reduces the burden of understanding in the mind of the maintainer. This paper explores a clustering algorithm based on the Hopfield neural net algorithm that clusters automatically using lexical similarity. Using that algorithm, this paper presents a tool that can aid the maintainer in maintaining a KBS. The tool is the Rule Base Clusterizer (RBC) which structures the KBS rule base to make it appear easy to understand for the maintainer. This paper shows by using entropy and Miller's number that the RBC finds the best clustering that produces both an adequate amount of information and acceptable sized clusters. The paper also presents three examples of running the RBC on real-world rule bases. © 1998 John Wiley & Sons, Ltd.