Semi-supervised Robust Dual-graph Concept Factorization via L2,1 Norm

Recently, as an effective paradigm for representation learning, Concept Factorization (CF) has attracted a great deal of interests in the areas of machine learning and data mining. However, it is difficult for existing CF related algorithms to effectively learn data manifold and feature manifold while utilizing the label information of data to learn more discriminative features. To overcome this problem, we propose a novel method called semi-supervised robust dual-graph CF (named RGCCF), which considers the manifold of data and features and combines the limited label information. Furthermore, an effective optimized method is proposed to solve the RGCCF problem. The experimential results on real-world datasets shows that RGCCF can achieve better clustering performance compared with the other CF-based methods.