【摘要】 针对基于会话的推荐算法(SBRS)在建模会话表示时,缺乏考虑会话中物品之间多元关联关系和用户重复性消费的行为模式,提出一种基于超图卷积网络的重复性消费会话推荐算法。算法首先根据用户的会话序列组建超图和线图,并通过超图卷积网络建模会话内物品之间多元关联关系和会话间交叉信息;接着通过注意力网络生成用户的意图表示;然后构建重复—探索模块以建模用户重复消费的行为模式;最后根据生成的会话表示预测下一个产生交互的物品评分,进行推荐。在2个公开的现实数据集上的大量实验结果表明,所提模型在召回率和平均倒数排名指标上优于其他基线算法。更多还原

【Abstract】 In this paper, a session-based recommendation based on hypergraph convolutional network is proposed to address the limitations of existing approaches in modeling the beyond pair-wise relations between items within a session and user behavior patterns of repeated consumption. Methodologically, first, the undirected hypergraphs and line graphs are generated based on all the sessions, then the embedding of the items are inputted into the hypergraph convolutional networks to capture the beyond pair-wise relations and the cross-session information, in addition,a repeat-explore module is employed to model the repeat consumption behavior patterns. Finally, according to the session representation, the next interaction will be predicted. Extensive experimental results on two real-world datasets demonstrate that the proposed model outperforms other baseline algorithms in terms of Recall and mean reciprocal rank(MRR).更多还原