Community detection method based on robust semi-supervised nonnegative matrix factorization

Abstract

Nonnegative Matrix Factorization (NMF) has been widely used to resolve the problem of community detection in complex networks. The present NMF-based methods for community detection cannot effectively integrate prior knowledge and deal with noises existing in complex networks, thus their performance still needs to be further improved. Aiming at these problems, we propose an approach for community detection based on robust semi-supervised NMF (RSSNMF). This method is able to combine must-link and cannot-link pairwise constraints based on semi-supervised NMF model and enhance the robustness from using the objective function based on ℓ2,1 norm. The community detection model of RSSNMF can be optimally solved by using the iterative update rules, of which the convergence can be strictly proved. Extensive comparative experiments have been conducted on four typical complex networks, and the results show that RSSNMF has better performance than other similar methods. Furthermore, RSSNMF is more robust and can reduce negative impacts from noises effectively on the performance of community detection.