Subset Infinite Relational Models

摘要: We propose a new probabilistic generative model for analyzing sparse and noisy pairwise relational data, such as friend-links on social network services customer records in online shops. Real-world data often include large portion of non-informative entries. Many existing stochastic blockmodels suffer from these irrelevant entries because their rather simpler forms priors. The proposed incorporates latent variable that explicitly indicates whether each entry is relevant or not to diminish bad effects associated with data. Through experiments using synthetic real sets, we show the can extract clusters stronger relations among within cluster than obtained by conventional model.