Signed Network Community Mining Based on Fine-grained Signed Stochastic Block Model

The reasoning of the stochastic block model mainly includes model learning and model selection, which respectively determine the block number K and the estimated parameters. The fine-grained stochastic block model (fg-SBM) can reduce the time complexity of parameter estimation, but it cannot be directly applied to signed networks. To this end, an improved model called fine-grained signed stochastic block model (fgs-SBM) is presented for signed network community mining. At the same time, in order to further reduce the time complexity of the signed network community mining algorithm, we choose to apply the block-wise SBM learning(BLOS) algorithm to integrate minimum message length (MML )and expectation maximization (EM) algorithm to realize parallel learning of parameter estimation and model selection. By comparing with the most advanced methods in synthetic networks and real networks, the advantages of this method in community mining of exploratory networks are illustrated.