Regularized asymmetric nonnegative matrix factorization for clustering in directed networks

摘要

In network analysis, clustering is a key task of dividing a network into logical and meaningful groupings of components. There are various methods to cluster nodes in undirected networks, however, little is known about clustering in directed networks. In this paper, we propose a regularized asymmetric nonnegative matrix factorization (RANMF) algorithm for clustering in directed networks. In a given directed network, the RANMF exploits the pairwise similarity of nodes to make close nodes belong to the same cluster under the guidance of prior information of the network. We also prove the convergence of the RANMF algorithm and provide real-world experiments to show its performance. The experimental results show the superiority of our RANMF algorithm in terms of several clustering validity indices. (C) 2019 Elsevier B.V. All rights reserved.