A novel method for rule extraction in a knowledge-based innovation tutoring system

Abstract

Symbolic, subsymbolic, and hybrid approaches to rule extraction have so far relied on subsets of first-order logic to cope with the expressiveness trade-off of knowledge representation, on black-box approaches based on artificial neural networks, or on frequent association rule mining in the knowledge discovery and data mining fields. In this article, we present an entirely new method for rule extraction in knowledge-based systems that consists in retrieving an initial set of rules extracted from a knowledge base using conventional logical approaches and then ranking this initial set of rules applying a psychologically motivated multicriteria decision analysis method. We show how this method can be used to implement a knowledge-based management system, demonstrate that this method outperforms the most efficient algorithms for rule extraction proposed to date in the knowledge representation and knowledge discovery fields, and describe its implementation in a knowledge-based innovation tutoring system.