Modularized tri-factor nonnegative matrix factorization for community detection enhancement

Community structure detection is a fundamental problem for understanding the relationship between the topology structures and the functions of complex networks. NMF-based models are a promising method for identifying communities from networks, but most of them require the number of communities in advance, which is inconvenient for real applications. Also, the basic NMF model could not reflect the characteristics of networks more comprehensively under the sole nonnegative constraint. In this paper, we develop a novel modularized tri-factor nonnegative matrix factorization model which combines the modularized information as the regularization term, leading to improved performance in community detection. Besides, we utilize general modularity density to determine the number of communities. Finally, the effectiveness of our approach is demonstrated on both synthetic and real-world networks. (C) 2019 Elsevier B.V. All rights reserved.