A new document clustering algorithm for topic discovering and labeling

Abstract

In this paper, we introduce a new clustering algorithm for obtaining labeled document clusters that accurately identify the topics of a text collection. In order to determine the topics, our approach relies on both probable term pairs generated from the collection and the estimation of the topic homogeneity associated to term pair clusters. Experimental results obtained over two benchmark text collections demonstrate the utility of this new approach. © 2008 Springer-Verlag Berlin Heidelberg.