A1302
Title: Variational Bayesian inference for bipartite MMSBM with applications to collaborative filtering
Authors: Panpan Zhang - Vanderbilt University Medical Center (United States) [presenting]
Abstract: Motivated by the connections between collaborative filtering and network clustering, a network-based approach is considered to improve rating prediction in recommender systems. A novel Bipartite Mixed-Membership Stochastic Block Model (BM2) with a conjugate prior from the exponential family is proposed. The analytical expression of the model is derived, and a variational Bayesian expectation-maximization algorithm is introduced, which is computationally feasible for approximating the untractable posterior distribution. Extensive simulations are carried out to show that BM2 provides a more accurate inference than standard SBM with the emergence of outliers. Finally, the proposed model is applied to a MovieLens dataset, and it is found that it outperforms other competing methods for collaborative filtering.
