The experimental evaluation of rules partitioning conception for knowledge base systems

Abstract

This article presents the theoretical background for implementation of KBExplorer software package and the summary of the empirical study focused on the evaluation of this software on large, real-word knowledge bases. KBExplorer package is the own, originally designed software which provides most of the expert system shell’s common functions. The fundamental part of such software is the KBExpertLib library. This library allows to build domain expert systems using Java programming language. The first part of experiments was focused on the effectiveness of rules partition algorithm and estimation of the memory occupancy for additional data necessary for storing information about rules groups. The effectiveness evaluation of the forward and backward inference algorithms was the main goal of the second part of the experiments.