Efficient Ensemble Methods for Document Clustering

Abstract

Recent ensemble clustering techniques have been shown to be effective in improving the accuracy and stability of standard clustering algorithms. However, an inherent drawback of these techniques is the computational cost of generating and combining multiple clusterings of the data. In this paper, we present an efficient kernel-based ensemble clustering method suitable for application to large, high-dimensional datasets such as text corpora. To decrease the time required to generate the ensemble members, we employ a prototype reduction scheme that makes use of a density-biased selection strategy to construct a smaller kernel matrix that represents a good proxy for the original data. Evaluations performed on text data demonstrate that this process leads to a significant decrease in running time, while maintaining high clustering accuracy.