Modeling Dynamic User Preference via Dictionary Learning for Sequential Recommendation

Capturing the dynamics in user preference is crucial to better predict future behaviors because preferences often drift over time. Many existing recommendation algorithms -- including both shallow and deep ones model such independently, i.e., static dynamic are not modeled under same latent space, which makes it difficult fuse them for recommendation. This paper considers problem of embedding a user's sequential behavior into space preferences, namely translating sequence preference. To this end, we formulate task as dictionary learning problem, learns: 1) shared matrix, each row represents partial signal across users; 2) posterior distribution estimator using autoregressive integrated with Gated Recurrent Unit (GRU), can select related rows represent conditioned on his/her past behaviors. Qualitative studies Netflix dataset demonstrate that proposed method capture drifts time quantitative multiple real-world datasets achieve higher accuracy compared state-of-the-art factorization neural methods. The code available at https://github.com/cchao0116/S2PNM-TKDE2021.