Unveiling the structure of multi-attributed networks via joint non-negative matrix factorization

Finding community structure (CS) has long been a core network science problem with various applications. Despite a vast amount of work on the problem, most current methods only focus on network topology, thus, perform poorly on real networks with labels and ground-truth communities. In this paper, we propose 3NCD (Network topology - Node attribute - NMF - Community Detection) algorithm that incorporates information from both network topology and node attributes in a unified non-negative matrix factorization (NMF) framework. The proposed algorithm not only achieves considerably higher accuracy but also runs markedly faster than the state-of-the-art methods. The superiority of our method is demonstrated in our experiments on three collections of real social networks with known ground-truth communities. Our experiments also suggest that the proposed method is robust against missing (incomplete) links and nodes´ information.