Discriminative semi-supervised non-negative matrix factorization for data clustering

Abstract

Recently semi-supervised non-negative matrix factorization (NMF) has received a lot of attentions in computer vision, information retrieval and pattern recognition, because that partial label information can produce considerable improvement in learning accuracy of the algorithms. However, the existing semi-supervised NMF algorithms cannot make full use of label information, that is, they cannot guarantee that the labeled data of different clusters will not be classified into a same group in the new representation space. In this paper, we propose a novel discriminative semi-supervised NMF (DSSNMF) algorithm, which utilizes the label information of a fraction of the data as a discriminative constraint. We explore the proposed DSSNMF method with two different cost function formulations and provide the corresponding update rules for the optimization problems. Empirical experiments demonstrate the effectiveness of our novel algorithm through a set of evaluations based on real-world applications.