Overlapping Community Recovery in Incomplete Graphs With Partial Prior Information

摘要: In this paper, we present an extension of our original KroMFac framework that deals with the problem of recoverying overlapping community structures from incomplete graphs with both missing nodes and edges. Unlike the original work where the number of communities is known, we consider community recovery without such prior information. To solve this problem, we treat the number of communities as a latent variable and uncover the community structures by solving a Bayesian non-negative matrix factorization (BNMF) optimization problem in terms of maximizing the likelihood of the underlying graph. We empirically show the superiority of our approach in terms of normalized mutual information (NMI) over two baseline schemes on synthetic graphs.