Multi-Behavior Sequential Recommendation with Low-Rank Decomposed Self-Attention
Zhaoju Zeng, Xiaodong Mu, Xuan Wei, Tao Jiang · 2022
Sequential recommendation seeks to simulate the changing preferences of users over time by using the sequential item information of user interactions. Most of the current sequential recommendation methods focus on a single row for sequential recommendation. In practical recommendation scenarios, user interactions are diverse, and it is necessary to mine multiple types of interaction data to predict user preferences over time. To this end, we propose a multi-behavior sequence recommendation method based on low-rank self-attentive decomposition networks. The dynamic interests of users are captured by exploring the behavioral relationships within and between multi-behavior sequences. Experiments on publicly available datasets show that our proposed approach consistently outperforms state-of-the-art recommendation models.