Mining mood-specific movie similarity with matrix factorization for context-aware recommendation

Context-aware recommendation seeks to improve recommendation performance by exploiting various information sources in addition to the conventional user-item matrix used by recommender systems. Recommendations should also usually strive to satisfy a specific purpose. Within the Moviepilot mood track of the context-aware movie recommendation challenge, we propose a novel movie similarity measure that is specific to the movie property demanded by the challenge, i.e., movie mood. Our measure is further exploited by a joint matrix factorization model for recommendation. We experimentally validate the effectiveness of the proposed algorithm in exploiting mood-specific movie similarity for the recommendation with respect to several evaluation metrics, demonstrating that it could outperform several state-of-the-art approaches. In particular, mood-specific movie similarity is demonstrated to be more beneficial than general mood-based movie similarity, for the purpose of mood-specific recommendation.