A Bayesian Framework for Learning Shared and Individual Subspaces from Multiple Data Sources

摘要: This paper presents a novel Bayesian formulation to exploit shared structures across multiple data sources, constructing foundations for effective mining and retrieval disparate domains. We jointly analyze diverse sources using unifying piece of metadata (textual tags). propose method based on Probabilistic Matrix Factorization (BPMF) which is able explicitly model the partial knowledge common datasets subspaces specific each dataset individual subspaces. For proposed model, we derive an efficient algorithm learning joint factorization Gibbs sampling. The effectiveness demonstrated by social media tasks single media. solution applicable wider context, providing formal framework suitable exploiting as well mutual present heterogeneous many kinds.