Fuzzy community detection via modularity guided membership-degree propagation

In complex network analysis, fuzzy community detection is a challenging task that aims to reveal the network structure by assigning each vertex quantitative membership-degrees to various communities. In this paper, we propose a fuzzy community detection method that iteratively propagates membership-degrees of all vertices. In each iteration, a candidate seed vertex of a potential community is first selected according to the topological characteristics. After that, the membership-degrees are propagated among adjacent vertices so that a number of communities can be obtained with respect to all selected seeds. To ensure that the modularity keeps improving, in each iteration we discard the selected seeds that decreases the modularity of the community decomposition. In this manner, the topological information about the network can be fully utilized, and communities gradually emerge along with the acceptance of new seeds. Experimental results on real-world and synthetic networks demonstrate that our approach has impressive performance and is robust on both disjoint and fuzzy community detections. Moreover, the proposed approach exhibits a high flexibility between computational complexity and overall performance.