Fast Co-clustering Using Matrix Decomposition

Co-clustering is a powerful data mining technique with varied applications such as text clustering, web-log mining and microarray analysis. Simultaneously clustering rows and columns (co-clustering) of large data matrix is an important problem with these wide applications. Current co-clustering techniques such as information-theoretic and Bayesian based methods provide good accuracy, but are computationally very expensive. Real data are noisy due to measurement technology limitation and experimental variability which prohibits co-clustering models from revealing true clusters corrupted by noise. Moreover, data matrices involving a large number of rows and columns limit their applicability. In this paper, we utilize correspondence analysis algorithm to process matrix decomposition and then make use of Bayesian approach for co-clustering. We find that utilizing the two methods synthetically is very significative to solve actual problems. Experiments on synthetic and real world data demonstrate the efficiency and effectiveness of our algorithm.