Multi-viewpoint clustering analysis

Knowledge-based systems have a wide commercial applicability. However, a credible validation methodology for knowledge-based systems is currently lacking. Our research addresses the feasibility of partitioning rule-based systems into a number of meaningful units to enhance the comprehensibility, maintainability and reliability of expert systems software. Preliminary results have shown that no single structuring principle or abstraction hierarchy is sufficient to understand complex knowledge bases. We therefore propose the Multi-Viewpoint - Clustering Analysis (MVP-CA) methodology to provide multiple views of the same expert system. MVP-CA provides an ability to discover significant structures within the rulebase by providing a mechanism to structure both hierarchically (from detail to abstract) and orthogonally (from different perspectives). Here we describe our approach to understanding large knowledge bases using MVP-CA. We demonstrate the need for MVP-CA using a couple of small classic rulebases, as well as a deployed knowledge-based system that navigates the Space Shuttle’s entry. We also discuss the impact of this approach on verification and validation of knowledge-based systems. MVP-CA provides an essential first step towards building an integrated environment for verification and validation of knowledge-based applications. It will allow one to build reliable knowledge-based systems by suitably abstracting, structuring, and otherwise clustering the knowledge in a manner which facilitates its understanding, manipulation, testing and utilization.