Databases for knowledge bases: empirical study of a knowledge base management system for a semantic network

Abstract

Database management is a mature technology. By contrast, management of knowledge bases poses a multitude of questions for which no generally accepted solutions appear to be available so far. The study presented here addresses the question whether knowledge base management systems (KBMS) can be understood as simply extending the functionality of database management system (DBMS), or whether more fundamental differences exist between the two. To better focus the question, an answer is attempted on the basis of established database technology such as relational systems. Since a close connection has already been established between the relational model and Horn logic or logic with fixpoint semantics, the paper concentrates on the narrower question of the support for knowledge representation formalisms such as semantic networks. The premise is that a substantive answer can currently be given only on the basis of empirical studies. Consequently, a careful analysis is presented of a case study that started out from an extended KL-ONE formalism as the knowledge base model for which a suitable set of operators was defined and the relational model as the supporting database technology. The implementation of the KBMS by means of a relational DBMS is retraced in detail, the various difficulties that arose during the mapping processes are pointed out as well as the compromises that had to be entered into. Finally, a resumé of the experiences is drawn in order to postulate some causes for the difficulties observed, and a number of suggestions and first experiences aiming at alleviating the suspected causes are presented.