Hypergraph temporal multi-behavior recommendation

As the scale of e-commerce and the number of item categories increase, user behaviors become increasingly diverse, and the real relationships between users and items in recommendation systems become considerably more complex. One of the emerging areas of research in this context is a multi-behavior recommendation, which aims to consider various types of user behavior to better predict user preferences by reflecting multiple behavior patterns. A primary challenge in current multi-behavior recommendation tasks is extracting user behavior temporality and behavior discrimination. Most existing studies cannot extract users’ temporal behavioral patterns and analyze the influence and relevance of various types of behaviors. To address this challenge, we propose a hypergraph temporal multi-behavior recommendation framework consisting of a temporal graph convolution network and a behavior-independent hypergraph. Temporal graph convolution network integrates a graph convolution network with a gated recurrent unit to extract the temporality and relationship of user–item interactions, and behavior independent hypergraph groups users and items with similar behavior patterns and analyzes high-order group relationships for user–item interactions. Our proposed framework can capture users’ temporal behavior dynamics and behavior discrimination by reflecting increasingly complex high-order relationships. We performed comparative experiments based on the hit ratio and normalized discounted cumulative gain metrics using three real-world e-commerce datasets and recorded superiority over the baseline model. This proves that the proposed model, hypergraph temporal multi-behavior recommendation, improves the ability to capture the temporality of user behaviors and effectively enhances the differentiation of each behavior.

Authors

• Jooweon Choi Chung-Ang University
• JuneHyoung Kwon Chung-Ang University
• Yeonghwa Kim
• YoungBin Kim Chung-Ang University

Keywords

• Multi-behavior recommendation
• Graph neural networks
• Hypergraph neural networks