A hybrid heuristic for overlapping community detection through the conductance minimization

Community structures, which are sets of elements that share some relationship between themselves, can be found in several real-world networks. Many of these communities, also known as clusters, can share elements, i.e., they may overlap. Identifying such overlapping clusters is usually a harder task than finding non-overlapping ones and, therefore, it needs more sophisticated methods. In this work we proposed a hybrid heuristic for detecting overlapping clusters in networks. An overlapping clustering is generated through the solving of a mixed-integer linear program using, as input, a heterogeneous set of good-quality clusters. This set is produced by two state-of-the-art overlapping community detection algorithms. In addition, some local search methods for conductance minimization are used to improve the quality of the clustering generate by our hybrid heuristic. Test results in artificial and real-world graphs show that our approach is able to detect overlapping clusters with better overall conductance than methods in the state of the art.