A supervised approach to community detection problem: how to improve Louvain algorithm by considering fuzzy measures

Abstract

Community detection problems are one of the most important problems in Social Network Analysis. Based on the Louvain algorithm, in this paper we propose a supervised technique to address the classic community detection problem in both directed and undirected networks. Our proposal is developed on the basis of extended fuzzy graphs, specifically paying attention to the notion of flow. We present a parametric and aggregation supervised approach that uses the flow capacity in terms of fuzzy information, in order to obtain realistic and global solutions, going one step further than local previous results. We evaluate the performance of that supervised technique by considering several benchmark and real-world networks. Taking into account the directed modularity, this new approach is developed under the machine learning paradigm, carrying through with two consecutive phases. The results obtained allow us to assert the goodness of our new supervised technique, beyond others existing algorithms