Modification to K-medoids and CLARA for effective document clustering

Abstract

Document clustering plays an important role in several applications. K-Medoids and CLARA are among the most notable algorithms for clustering. These algorithms together with their relatives have been employed widely in clustering problems. In this paper we present a solution to improve the original K-Medoids and CLARA by making change in the way they assign objects to clusters. Experimental results on various document datasets using three distance measures have shown that the approach helps enhance the clustering outcomes substantially as demonstrated by three quality metrics, i.e. Entropy, Purity and F-Measure.