Label propagation based semi-supervised non-negative matrix factorization for feature extraction

As a feature extraction method, Non-negative Matrix Factorization (NMF) has attracted much attention due to its effective application to data classification and clustering tasks. In this paper, a novel algorithm named Label propagation based Semi-supervised Non-negative Matrix Factorization (LpSNMF) is proposed. For the sake of making full use of label information, our LpSNMF algorithm takes the distribution relationships between the labeled and unlabeled data samples into consideration and integrates the procedures of class label propagation and matrix factorization into a joint framework. Moreover, an iterative updating optimization scheme is developed to solve the objective function of the proposed LpSNMF and the convergence of our scheme is also proven. Extensive experimental results on several UCI benchmark data sets and four image data sets (such as Yale, CMU PIE, UMIST, and COIL20) demonstrate that by propagating the label information and factorizing the matrix alternately, our algorithm can obtain better performance than some other algorithms.