联合对比学习的图神经网络会话推荐

Graph Neural Network Session Recommendation Assisted with Contrastive Learning

基于会话的推荐(SBR)是一项具有挑战性的任务,其目的是根据匿名行为序列推荐项目。本文提出了一种新的方法,称为联合对比学习的图神经网络会话推荐(CLGNN),在图注意力机制的基础上,用对比学习辅助训练,以获得更好的推荐结果。具体来说,CLGNN首先在会话图上采用注意力机制学习项目嵌入,然后聚合会话内的项目生成会话嵌入,最后使用会话嵌入和候选项目嵌入计算分数生成推荐,同时使用对比学习优化项目嵌入空间。以几种常见的评价指标为依据,在真实的两个数据集上进行实验,结果表明本文模型推荐性能良好。

Session based recommendation(SBR)is a challenging task aimed at recommending items based on anonymous behavior sequences.This article proposes a new method called Graph Neural Network Session Recommendation Assisted with Contrastive Learning(CLGNN).Based on the graph attention mechanism,contrastive learning is used to assist training in order to obtain better recommended results.Specifically,CLGNN uses attention mechanism to learn item embedding on the session graph,and then aggregates items within the session to generate session embedding.Finally,generate recommendations using session embedding and candidate item embedding to calculate scores.At the same time,contrastive learning is used to optimize item embedding space.Based on several common evaluation index methods,experiments were conducted on two real datasets,and the results showed that the model recommendation performance in this article is good.