Different methods for cluster’s representation and their impact on the effectiveness of searching through such a structure

Abstract

In this paper the topic of clustering and searching through clusters generated from real-world knowledge bases is discussed. Authors analyze three methods of cluster’s representatives creation, focusing on their advantages and flaws. What is more the authors introduce a concept of forward-chaining inference which uses a density-based DBSCAN algorithm to generate cluster of rules and speed up the whole inference process. The experiments were conducted on real-world knowledge bases with a relatively large number of rules to evaluate the efficiency of the proposed approach.