Fuzzy clustering in community detection based on nonnegative matrix factorization with two novel evaluation criteria

Clustering or community detection is one of the most important problems in social network analysis, and because of the existence of overlapping clusters, fuzzy clustering is a suitable way to cluster these networks. In fuzzy clustering, in addition to the correctness of the clusters assigned to each node, the produced membership of one node to each cluster is also important. In this paper, we introduce a new fuzzy clustering algorithm based on the nonnegative matrix factorization (NMF) method. Despite the well-known fuzzy clustering techniques like FCM, the proposed method does not depend on any parameter. Also, it can produce appropriate memberships based on the network structure and so identify the overlap nodes from non-overlap nodes, well. Also, to evaluate the validity of such fuzzy clustering algorithms, we propose two new evaluation criteria (SFEC and UFEC), which are constructed based on the neighborhood structure of nodes and can evaluate the memberships. Experimental results on some realworld networks and also many artificial networks show the effectiveness and reliability of our proposed criteria. (C) 2017 Elsevier B.V. All rights reserved.