A Node Influence Based Memetic Algorithm for Community Detection in Complex Networks

Abstract

Community structure is a significant property when analyzing the features and functions of complex systems. Heuristic algorithm-based community detection treats finding the community structure as an optimization problem, which has received great attentions in a variety of fields these years. Several community detection methods have been proposed. To make an approach of detecting the community structure in a more efficient way, a node influence based memetic algorithm (NIMA), considering node influence, is proposed in this paper. The NIMA consists of three main parts. First of all, a transition probability matrix-based initialization is employed to accelerate the convergence speed and provide an initial population with great diversity. Secondly, a network-specific crossover and a node degree-based mutation are designed to enlarge the search space and keep effective information. Last, a multi-level greedy search is deployed to find the potential optimal solutions quickly and effectively. Extensive experiments on 28 synthetic and 6 real-world networks demonstrate that compared with 11 existing algorithms, the proposed NIMA has effective performance on detecting communities in complex networks.