摘 要：社交推荐旨在将社交信息与用户 / 项目交互结合起来，以缓解评分预测任务中的数据稀疏问题。针对现有的方法都局限于局部特征的权重分配和处理，忽略了全局特征在社交图中的重要性，因此，提出通过利用多头自注意力机制在图神经网络框架下关注全局特征的权重分配的基于图注意力机制优化的社交推荐模型，从而提高推荐效果。在 Ciao 和 Epinions 两个数据集上验证表明，与现有相关基线相比，模型实现了更好的效果。
关键词：推荐系统；社交关系；注意力机制；图神经网络
DOI:10.19850/j.cnki.2096-4706.2022.17.026
中图分类号：TP391 文献标识码：A 文章编号：2096-4706（2022）17-0101-04
Research on Social Recommendation Algorithm Based on Graph Attention Mechanism Optimization
XU Feng
(Anhui University of Science and Technology, Huainan 232001, China)
Abstract: Social recommendation aims to combine social information with user/item interactions to alleviate the data sparse problem in score prediction tasks. In views of existing methods are all limited to the weight assignment and processing of local features, ignoring the importance of global features in social graphs. Therefore, this paper proposes a social recommendation model through utilizing the multi-head self-attention mechanism, which is under the graph neural network framework, focuses on the weight assignment of global features and based on a graph attention mechanism optimization, so as to improve the recommendation effect. The validation on two datasets, Ciao and Epinions, shows that the model achieves better effect compared to existing related baselines.
Keywords: recommendation system; social relationship; attention mechanism; graph neural network
参考文献：
[1] CHEN J J，XIN X，LIANG X F，et al. GDSRec：GraphBased DecentralizedCollaborative Filtering for SocialRecommendation [J].IEEE Transactions on Knowledge and Data Engineering，2022：1.
[2] 潘明玥 . 基于图注意力网络的社交推荐系统研究 [D]. 长春：东北师范大学，2021.
[3] 王英博，孙永荻 . 基于 GNN 的矩阵分解推荐算法 [J]. 计算机工程与应用，2021，57（19）：129-134.
[4] FAN W Q，LI Q，CHENG M. Deep modeling of social relations for recommendation [EB/OL].[2022-06-17].https://dl.acm.org/ doi/pdf/10.5555/3504035.3505064.
[5] FAN W Q，MA Y，LI Q，et al. Graph neural networks for social recommendation [J/OL].arXiv：1902.07243[cs.IR].[2022-06-17]. https://arxiv.org/abs/1902.07243.
[6] SONG W P，XIAO Z P，WANG Y F，et al. Session-based Social Recommendation via Dynamic Graph Attention Networks [J/OL].arXiv：1902.09362 [cs.IR].[2022-06-18].https://arxiv.org/ abs/1902.09362.
[7] SALAKHUTDINOV R，MNIH A. Probabilistic matrix factorization [EB/OL].[2022-06-18] https://www.doc88.com/ p-6961593834520.html.
[8] JAMAlI M，ESTER M. A matrix factorization technique with trust propagation for recommendation in social networks [C]// Proceedings of the fourth ACM conference on Recommender systems. Barcelona：Association for Computing Machinery，2010：135-142.
作者简介：许峰（1996—），男，汉族，安徽合肥人，硕士研究生在读，研究方向：推荐系统。