A New Information Theory Based Clustering Fusion Method for Multi-view Representations of Text Documents

Abstract

Multi-view clustering is a complex problem that consists in extracting partitions from multiple representations of the same objects. In text mining and natural language processing, such views may come in the form of word frequencies, topic based representations and many other possible encoding forms coming from various vector space model algorithms. From there, in this paper we propose a clustering fusion algorithm that takes clustering results acquired from multiple vector space models of given documents, and merges them into a single partition. Our fusion method relies on an information theory model based on Kolmogorov complexity that was previously used for collaborative clustering applications. We apply our algorithm to different text corpuses frequently used in the literature with results that we find to be very satisfying.