DDOC: Overlapping Clustering of Words for Document Classification

摘要

In this paper we study the interest of integration of an overlapping clustering approach rather than traditional hard-clustering ones, in the context of dimensionality reduction of the description space for document classification. The Distributional Divisive Overlapping Clustering (DDOC) method is briefly presented and compared to Agglomerative Distributional Clustering (ADC) and Information-Theoretical Divisive Clustering (ITDC) on the two corpus Reuters-21578 and 20Newsgroup.