Recurrent Attentive Neural Networks for Sequential Recommendation
Lei Tan, Jinmao Xu, Daofu Gong, Fenlin Liu · 2023
Sequential recommendation is essential in modern online service platforms. By modeling the evolving preferences of a user from the historical behavior sequence, sequential recommendation aims to predict the next interaction of the user in the near future. For sequential recommendation, it is challenging to comprehensively characterize user preferences based on historical behavior sequences. In this paper, we propose a Recurrent Attentive Neural Networks model (RANN) for sequential recommendation, which characterizes the user preference from the long-term preference, short-term interest, and current sequential pattern. Specifically, the long short-term memory network is employed to generate the cell state and hidden state at each time step for each user, which represent the long-term preference and the current sequential pattern of the user, respectively. And the scaled dot-product attention is utilized to capture the short-term interest feature among the most recent hidden states. Finally, user preference features and item embeddings are integrated into a next-item prediction network for sequential recommendation. Experimental results on the real-world dataset verify the superiority of RANN against state-of-the-art baselines.