Semi-supervised concept factorization for document clustering

Abstract

Nonnegative Matrix Factorization (NMF) and Concept Factorization (CF) are two popular methods for finding the low-rank approximation of nonnegative matrix. Different from NMF, CF can be applied not only to the matrix containing negative values but also to the kernel space. Based on NMF and CF, many methods, such as Graph regularized Nonnegative Matrix Factorization (GNMF) and Locally Consistent Clustering Factorization (LCCF) can significantly improve the performance of clustering. Unfortunately, these are unsupervised learning methods. In order to enhance the clustering performance with the supervisory information, a Semi-Supervised Concept Factorization (SSCF) is proposed in this paper by incorporating the pairwise constraints into CF as the reward and penalty terms, which can guarantee that the data points belonging to a cluster in the original space are still in the same cluster in the transformed space. By comparing with the state-of-the-arts algorithms (KM, NMF, CF, LCCF, GNMF, PCCF), experimental results on document clustering show that the proposed algorithm has better performance in terms of accuracy and mutual information.