Mixed Membership Estimation for Weighted Networks

Abstract: Community detection in overlapping un-weighted networks in which nodes can belong to multiple communities is one of the most popular topics in modern network science during the last decade. However, community detection in overlapping weighted networks in which edge weights can be any real values remains a challenge. In this article, to model overlapping weighted networks with latent community memberships, we propose a generative model called the degree-corrected mixed membership distribution-free model which can be viewed as generalizing several previous models. First, we address the community membership estimation of the proposed model by an application of a spectral algorithm and establish a theoretical guarantee of consistency. We then propose overlapping weighted modularity to measure the quality of overlapping community detection for weighted networks with positive and negative edge weights. To determine the number of communities for weighted networks, we incorporate the algorithm into the overlapping weighted modularity. We demonstrate the advantages of degree-corrected mixed membership distribution-free model and overlapping weighted modularity with applications to simulated data and eleven real-world networks.