Mining Clustering Dimensions

Abstract

Although it is common practice to produce only a single clustering of a dataset, in many cases text documents can be clustered along different dimensions. Unfortunately, not only do traditional clustering algorithms fail to produce multiple clusterings of a dataset, the only clustering they produce may not be the one that the user desires. To address this major limitation, we propose a novel clustering algorithm for inducing multiple clusterings along the important dimensions of a dataset. Its ability to reveal the important clustering dimensions of a dataset in an un-supervised manner is particularly appealing for those users who have no idea of how a data can possibly be clustered. We demonstrate its viability on several challenging text classification tasks.