Capturing the temporal dynamics of user preferences over items is importantfor recommendation. Existing methods mainly assume that all time steps inuser-item interaction history are equally relevant to recommendation, whichhowever does not apply in real-world scenarios where user-item interactions canoften happen accidentally. More importantly, they learn user and item dynamicsseparately, thus failing to capture their joint effects on user-iteminteractions. To better model user and item dynamics, we present theInteracting Attention-gated Recurrent Network (IARN) which adopts the attentionmodel to measure the relevance of each time step. In particular, we propose anovel attention scheme to learn the attention scores of user and item historyin an interacting way, thus to account for the dependencies between user anditem dynamics in shaping user-item interactions. By doing so, IARN canselectively memorize different time steps of a user's history when predictingher preferences over different items. Our model can therefore providemeaningful interpretations for recommendation results, which could be furtherenhanced by auxiliary features. Extensive validation on real-world datasetsshows that IARN consistently outperforms state-of-the-art methods.
展开▼
