Factorization Models for Context-/Time-Aware Movie Recommendations

In the scope of the Challenge on Context-aware Movie Recommendation (CAMRa2010), context can mean temporal context (Task 1), mood (Task 2), or social context (Task 3). We suggest to use Pairwise Interaction Tensor Factorization (PITF), a method used for personalized tag recommendation, to model the temporal (week) context in Task 1 of the challenge. We also present an extended version of PITF that handles the week context in a smoother way. In the experiments, we compare PITF against different item recommendation baselines that do not take context into account, and a non-personalized context-aware baseline.