Community detection using Ant Colony Optimization

Many complex networks have been shown to have community structure. How to detect the communities is of great importance for understanding the organization and function of networks. Due to its NP-hard property, this problem is difficult to solve. In this paper, we propose an Ant Colony Optimization (ACO) approach to address the community detection problem by maximizing the modularity measure. Our algorithm follows the scheme of max-min ant system, and has some new features to accommodate the characteristics of complex networks. First, the solutions take the form of a locus-based adjacency representation, in which the communities are coded as connected components of a graph. Second, the structural information is incorporated into ACO, and we propose a new kind of heuristic based on the correlation between vertices. Experimental results obtained from tests on the LFR benchmark and four real-life networks demonstrate that our algorithm can improve the modularity value, and also can successfully detect the community structure.