Time-Sensitive Collaborative Filtering through Adaptive Matrix Completion

Real-world Recommender Systems are often facing drifts in users’ preferences and shifts in items’ perception or use. Traditional stateof-the-art methods based on matrix factorization are not originally designed to cope with these dynamic and time-varying effects and, indeed, could perform rather poorly if there is no ”reactive”, on-line model update. In this paper, we propose a new incremental matrix completion method, that automatically allows the factors related to both users and items to adapt “on-line” to such drifts. Model updates are based on a temporal regularization, ensuring smoothness and consistency over time, while leading to very efficient, easily scalable algebraic computations. Several experiments on real-world data sets show that these adaptation mechanisms significantly improve the quality of recommendations compared to the static setting and other standard on-line adaptive algorithms.