Journal of Frontiers of Computer Science and Technology ›› 2023, Vol. 17 ›› Issue (4): 826-836.DOI: 10.3778/j.issn.1673-9418.2107009

• Theory·Algorithm • Previous Articles     Next Articles

Bilinear Diffusion Graph Recommendation Model Fusing User Social Relations

ZHU Ji, XIAO Xiaoli, YIN Bo, SUN Qian, TAN Dong   

  1. School of Computer and Communication Engineering, Changsha University of Science and Technology, Changsha 410114, China
  • Online:2023-04-01 Published:2023-04-01



  1. 长沙理工大学 计算机与通信工程学院,长沙 410114

Abstract: In order to solve the sparse problem of user behavior data in traditional collaborative filtering algorithms, this paper proposes a bilinear diffusion graph recommendation model that fuses user social relationships based on graph convolutional networks, and designs a bilinear diffusion aggregator. The aggregator consists of two core parts. One is the diffusion aggregation part which is used to capture user-neighbor interaction information. This part aggregates neighbor information from the local neighborhood of the social graph by modeling the dynamic diffusion process of the user’s social influence. The target user node is used to enrich the target user representation. The second is the bilinear aggregation part which is used to capture neighbor-neighbor interaction information. This part exploits the potential social interaction between the neighbors of the same level users, uses the inner product operation to highlight the common features in the neighborhood social information, and uses it as the auxiliary information of the target user to improve user embedding. In order to verify the effectiveness of the model, this paper conducts recommendation experiments on Yelp and Flickr datasets, and conducts an experimental comparison analysis with the existing recommendation models. Experimental results show that this model has a higher hit ratio and normalized discounted cumulative gain than the existing recommendation models. Therefore, the bilinear diffusion graph recommendation model that integrates users’ social relationships can effectively alleviate the sparseness of user behavior data and greatly improve the accuracy of recommendation.

Key words: social recommendation, bilinear network, graph convolutional network, feature fusion

摘要: 为了解决传统协同过滤算法中用户行为数据的稀疏问题,在图卷积网络的基础上,提出了一个融合用户社会关系的双线性扩散图推荐模型,并设计了双线性扩散聚合器。该聚合器由两大核心部分组成:一是用来捕捉用户-邻居交互信息的扩散聚合部分,该部分通过建模用户社交影响的动态扩散过程,从社交图的局部邻域中聚合邻居信息到目标用户节点上,以此丰富目标用户表征;二是用来捕捉邻居-邻居交互信息的双线性聚合部分,该部分通过挖掘同阶用户邻居之间潜在的社会交互,并使用内积操作突出邻居社交信息中的共有特征,将其作为目标用户的辅助信息完善用户嵌入。为了验证该模型的有效性,在Yelp和Flickr数据集上进行推荐实验,并与现有的推荐模型进行实验对照分析。实验结果显示,该模型较现有的推荐模型有更高的命中率和归一化折损累计增益。因此,融合用户社会关系的双线性扩散图推荐模型能够有效缓解用户行为数据的稀疏问题,并使得推荐准确率有了较大提升。

关键词: 社会化推荐, 双线性网络, 图卷积网络, 特征融合