计算机工程与应用 ›› 2020, Vol. 56 ›› Issue (24): 169-174.DOI: 10.3778/j.issn.1002-8331.1909-0339

• 模式识别与人工智能 • 上一篇    下一篇

融合社交网络用户潜在因子的社会化推荐

赵亮,陈平华,廖威平   

  1. 广东工业大学 计算机学院,广州 510006
  • 出版日期:2020-12-15 发布日期:2020-12-15

Social Recommendation Based on Latent Factors of Social Network Users

ZHAO Liang, CHEN Pinghua, LIAO Weiping   

  1. School of Computer, Guangdong University of Technology, Guangzhou 510006, China
  • Online:2020-12-15 Published:2020-12-15

摘要:

针对传统社会化推荐准确率不高的问题,提出一种融合社交网络用户潜在因子的推荐算法SGCN-MF。SGCN-MF考虑社交网络中用户的隐语义信息对推荐结果的影响。使用图卷积神经网络将用户-项目历史交互信息和用户社交网络进行编码嵌入,学习得到具有用户特征和网络结构信息的节点在低维向量空间的潜在特征表达;将用户潜在因子融入基于矩阵分解的社会化推荐模型中;使用梯度下降算法训练模型参数。在Filmtrust、Ciao和Epinions数据集上的实验表明,与传统的社会化推荐算法相比,SGCN-MF能够提升推荐的准确率。

关键词: 社会化推荐, 矩阵分解, 潜在因子, 社交网络, 图卷积神经网络

Abstract:

In order to solve the problem of the accuracy of social recommendation, a recommendation algorithm named SGCN-MF is proposed that fuses the latent factors of the user’s social network. The algorithm involves the influence of the implicit semantic information of the user in the social network on the result. user-project history information and user social networks are embedded using a graph convolutional neural network, then the user latent factors are integrated into the socialized recommendation model based on matrix decomposition, and finally the model parameters are trained by the gradient descent algorithm. Experiments on the Filmtrust, Ciao and Epinions dataset show that the algorithm can improve the accuracy of the recommendation results compared with the traditional social recommendation algorithm.

Key words: social recommendation, matrix factorization, latent factors, social networks, graph convolutional neural network