MGR: Metric Learning with Graph Neural Networks for Multi-behavior Recommendation

Traditional recommendation methods often suffer from the problems of sparsity and cold start. Therefore, researchers usually leverage Knowledge Graph as a kind of side information to alleviate these issues and improve the accuracy of recommendation results. However, most existing studies focus on modeling the single behavior of user-item interactions, ignoring the active effects of the multi-type behavior information in the recommendation performance. In view of this, we propose Metric Learning with Graph Neural Networks for Multi-behavior Recommendation (MGR), a novel sequential recommendation framework that considers both temporal dynamics and semantic information. Specifically, the temporal encoding strategy is used to model dynamic user preferences. In addition, the Graph Neural Network is utilized to capture the information from high-order nodes so as to mine the semantic description in multi-behavior interactions. Finally, symmetric metric learning helps to sort the item list to accomplish the Top-K recommendation task. Extensive experiments in three real-world datasets demonstrate that MGR outperforms the state-of-the-art recommendation methods.