Repeat Consumption Recommendation Based on Users Preference Dynamics and Side Information

We present a Coupled Tensor Factorization model to recommend items with repeat consumption over time. We introduce a measure that captures the rate with which the preferences of each user shift over time. Repeat consumption recommendations are generated based on factorizing the coupled tensor, by weighting the importance of past user preferences according to the captured rate. We also propose a variant, where the diversity of the side information is taken into account, by higher weighting users that have more rare side information. Our experiments with real-world datasets from last.fm and MovieLens demonstrate that the proposed models outperform several baselines.