A Fast Spectral Method to Solve Document Cluster Ensemble Problem

Abstract

The critical problem in cluster ensemble is how to combine clusterers to yield a final superior clustering result. In this paper, we introduce a spectral method to solve document cluster ensemble problem. Since spectral clustering inevitably needs to compute the eigenvalues and eigenvectors of a matrix, for large scale document datasets, itpsilas computationally intractable. By using algebraic transformation to similarity matrix we get a feasible algorithm. Experiments on TREC and Reuters document sets show that our spectral algorithm yields better clustering results than other typical cluster ensemble techniques without high computational cost.