关键词: 推荐系统, 图神经网络, 会话推荐, 对比学习, 图注意力网络

Abstract: To address the problem that session recommendation models ignore temporal information and spatial relationships between items, leading to an inability to accurately capture complex transition patterns, session-based recommendation model based on time-aware and space-enhanced dual channel graph neural network was proposed. First, for the temporal channel, adaptive temporal weights were used to process items, constructing a time-aware session graph to capture users' interest-shifting patterns through a time-aware graph neural network. Second, for the spatial channel, spatial relationships among items were embedded into a graph attention network to aggregate information from the perspective of spatial graph structures. Additionally, a contrastive learning strategy was introduced to enhance recommendation performance. Experimental results show that the y precision(P) and mean reciprocal rank (MRR) on three publicly available datasets, Diginetica, Tmall, and Nowplaying, achieve superior performance compared to the latest session recommendation models. Compared to the suboptimal models Atten-Mixer and GCE-GNN, P@10 improves by 2.09%, 24.97%, and 10.45%, while MRR@10 improves by 2.52%, 11.60%, and 4.43%, respectively, demonstrating the effectiveness of the proposed model.

Key words: recommendation system, graph neural network, session-based recommendation, contrastive learning, graph attention network