Community detection based on the Matthew effect

Abstract

Abstract Community structure exists in most real-world networks, such as social networks, smart grids, and transportation networks. Established approaches for community detection usually depend on some user-defined criteria (e.g., minimum cut, normalized cut, modularity, etc.). These criteria-based methods usually involve some optimization procedures and need to specify some parameters, which are thus time consuming and sensitive to parameters. In this paper, inspired by the Mathew effect of human society, we view a network as a social system and design a new algorithm called CDME (community detection based on the Matthew effect). Relying on the new concept, CDME has many desirable properties. It allows uncovering high-quality communities driven by dynamic CDME is also parameter free. More importantly, since CDME works in a local way and only needs to calculate the attractiveness of neighboring nodes, which lend itself to handling large-scale networks. Experiments on both synthetic and real-world data sets have demonstrated that CDME has many benefits and outperforms many state-of-the-art algorithms.