An entropy-based evaluation function for conceptual clustering

There are two important tasks in conceptual clustering: (1) successfully grouping objects which are closely related to each other into the same concept; and (2) deciding automatically the number of concepts for the given objects. In this paper, we propose an entropy-based function which is sensitive to the distribution of objects to evaluate the clustering quality. Based on the proposed function, we present a CLUSTER/2-like clustering system which produces clusters containing closely related objects and decides the number of clusters reasonably and automatically. Tests on some benchmarks are compared with respect to our approach and CLUSTER/2. From the experiment results, our system performs better than CLUSTER/2.