Discovering communities can promote the understanding of the structure, function and evolution in various systems. Overlapping community detection in poly-relational networks has gained much more interests in recent years, due to the fact that poly-relational networks and communities with pervasive overlap are prevalent in the real world. A plethora of methods detect communities from the poly-relational network by converting it to mono-relational networks first. Nevertheless, they commonly assume different relations are independent from each other, which is obviously unreal to real-life cases. In this paper, we attempt to relax this strong assumption by introducing a novel co-ranking framework, named MutuRank. It makes full use of the mutual influence between relations and actors to transform the poly-relational network to the mono-relational network. We then present a novel GMM-NK (Gaussian Mixture Model with Neighbor Knowledge) algorithm incorporating the impact from neighbors into the traditional GMM. Experimental results both on synthetic networks and the real-world network have verified the effectiveness of MutuRank and GMM-NK.
