重庆理工大学学报(自然科学) ›› 2023, Vol. 37 ›› Issue (8): 212-221.

• 智能技术 • 上一篇    下一篇

双通道超图卷积网络团购推荐

冯英杰,张志鸿,贾玉祥   

  1. (郑州大学 计算机与人工智能学院,郑州 450001)
  • 出版日期:2023-09-15 发布日期:2023-09-15
  • 作者简介:冯英杰,男,硕士研究生,主要从事数据挖掘、推荐算法研究,Email:zzufyj01@163.com;通信作者 贾玉祥,男,博 士,副教授,主要从事自然语言处理研究,Email:ieyxjia@zzu.edu.cn

Group-buying recommendation based on dual channel hypergraph convolutional network

  • Online:2023-09-15 Published:2023-09-15

摘要: 在社会化电子商务团购中,如何充分挖掘用户与参与者的潜在偏好是影响团购成 功的重要因素。现有的社会推荐模型大都利用成对关系挖掘用户的潜在偏好,未考虑用户和参 与者之间的复杂交互,无法在团购场景下建立用户的高阶关系。因此,提出了双通道超图卷积 网络(dualchannelhypergraphconvolutionalnetworkforgroupbuyingrecommendation,HCGR),并 根据团购中角色的不同,在通道内设置不同的消息传递机制,充分利用高阶社会关系更全面地 捕获用户潜在偏好。然后使用门控机制自适应聚合不同通道的用户嵌入信息,最终生成推荐结 果。在真实团购数据集上的实验表明,HCGR优于所有对比模型,总体性能提升 3.18%~ 446%。消融实验以及数据稀疏性实验进一步验证了模型的合理性。

关键词: 团购推荐, 社交网络, 超图, 双通道, 卷积网络

Abstract: In social e-commerce group buying, how to fully tap the potential preferences of users and participants is an important factor affecting the success of group buying. Most of the existing social recommendation models use pair-wise relationships to mine users’ potential preferences. However, considering the complex interaction between users and participants, it is necessary to establish a high-order relationship between users in the group buying scenario. Therefore, we propose a dual-channel Hypergraph Convolutional network for Group-buying Recommendation, HCGR, and set up different message passing mechanisms in the channel according to different roles in group buying, making full use of high-order social relations to capture users’ potential preferences more comprehensively. Then, the gating mechanism is used to adaptively aggregate the user embedding information of different channels, and finally generate the recommendation results. Experiments on real group buying datasets show that HCGR is superior to all comparison models, and the overall performance gets improved by 3.18%~4.46%. Ablation experiments and data sparsity experiments further verify the rationality of the proposed model.

中图分类号: 

  • TP391