KNOWLEDGE REPRESENTATION IN EXPERT SYSTEMS

Abstract

Knowledge representation is frequently said to be the central issue in expert systems research. The definition and special characteristics of an expert system demand that three criteria for knowledge representation be met: expressive adequacy, the ability to represent the necessary distinctions of the domain in the representation; explicitness, the accessibility of all necessary naturalness, the case with which the representation captures the structure and content of knowledge in the domain. To date, most expert systems have relied on one of two knowledge representation formalisms, rule-based production systems or frame-based conceptual graphs. A survey of the theoretical literature shows that both these formalisms are considered appropriate representations to meet the three criteria. An examination of systems utilizing these formalisms indicates that this is not the case. Important discrepancies exist between these theoretical arguments and the existing successful expert systems. In order to investigate these issues further an expert system for the interpretation of personality inventories in clinical psychology was designed and two prototypes were implemented. Based on the examination of the literature and the discussion of the prototypes the definitions of the three criteria appearing in the literature are seen to be inadequate. More precise definitions, taking into account the field of expert systems and other areas of research into knowledge representation, are proposed and issues crucial to the evaluation of any representation are raised.