Méthodes de vérification de bases de connaissances
Philippe Lafon · HAL (Le Centre pour la Communication Scientifique Directe) · 1991
In this work, we present our approach for automatic checking of rule based systems. Atfer an overview of the related works, this report consists of two distincts parts. The first one deals with the consistency of rules in Knowledge Bases (KB) using the attribute-value formalism with forward chaining. We present the MELOMIDIA system, a consistency checker looking for the specifications of Initial Fact Bases (IFB) such that the deductive closure of IFB and the rule base contains contradictory facts. In the same process, MELOMIDIA also detects the redundant or unfireable rules of the KB. The second one considers the problem of verification of first-order Rule Bases (RB). Our approach consists of matching two kinds of knowledge: on one hand, the KB to be studied, on the other hand, specifications of properties of some predicates used in the KB. This latter core of knowledge is some kind of implicit knowledge, which must be acquired from the KB designer. This is made possible through a descriptive model of predicate we designed for this task. Going further, this model allows us to tackle some methodological aspects of the development of KB Systems.