A local fuzzy method based on “p-strong” community for detecting communities in networks

Abstract

In this paper, we propose a local fuzzy method based on the idea of "p-strong" community to detect the disjoint and overlapping communities in networks. In the method, a refined agglomeration rule is designed for agglomerating nodes into local communities, and the overlapping nodes are detected based on the idea of making each community strong. We propose a contribution coefficient b c i v to measure the contribution of an overlapping node to each of its belonging communities, and the fuzzy coefficients of the overlapping node can be obtained by normalizing the b c i v to all its belonging communities. The running time of our method is analyzed and varies linearly with network size. We investigate our method on the computergenerated networks and real networks. The testing results indicate that the accuracy of our method in detecting disjoint communities is higher than those of the existing local methods and our method is efficient for detecting the overlapping nodes with fuzzy coefficients. Furthermore, the local optimizing scheme used in our method allows us to partly solve the resolution problem of the global modularity.

Link prediction in complex networks via modularity-based belief propagation

“…For example, the hierarchical random graph based link prediction framework with the assumption of hierarchical organization of networks tends to have high prediction accuracy if the observed network organizes hierarchically. [18] Similarly, if the observed network is organized into blocks [32,33] or distinct roles, the framework based on the stochastic block model (SBM) [34] will provide accurate link predictions. [19] Inspired by this fact, Wang et al viewed link addition in network evolution as a kind of link prediction algorithm and employed likelihood analysis to judge which mechanism is better among a given series of network formation mechanisms for explaining the evolution of a network, [28] while Zhang et al applied the likelihood analysis and link prediction methods to quantitatively measure how multiple mechanisms contribute to a network formation.…”