A weakly-supervised factorization method with dynamic graph embedding

Nonnegative matrix factorization (NMF) is an effective method to learn a vigorous representation of nonnegative data and has been successfully applied in different machine learning tasks. Using NMF in semi-supervised classification problems, its factors are the label matrix and the membership values of data points. In this paper, a dynamic weakly supervised factorization is proposed to learn a classifier using NMF framework and partially supervised data. Also, a label propagation mechanism is used to initialize the label matrix factor of NMF. Besides a graph based method is used to dynamically update the partially labeled data in each iteration. This mechanism leads to enriching the supervised information in each iteration and consequently improves the classification performance. Several experiments were performed to evaluate the performance of the proposed method and the results show its superiority compared to a state-of-the-art method.