| 摘 要: 针对如何在图卷积网络中融入用户的社交网络以及有效实现异构关系学习的问题,提出了一种包含隐式信任和影响的新颖的异构图卷积网络框架(HGCNTI)。该框架基于用户-用户二分图构建信任子图和影响子图,充分利用用户间的隐式关系达到增强用户-项目表示的目的;此外,设计了一个多视角元网络,从不同用户或项目中提取个性化信息,实现个性化知识转换的自适应增强。实验结果表明,在Ciao和Epinions两个数据集上,HGCNTI均表现出色。与各种最新基线相比,在Ciao数据集上,其召回率@5提升了22.6%,召回率@10提升了19.7%,NDCG@10提升了19%;在Epinions数据集上,NDCG@5提升了2.9%,精确率@10提升了4.5%。 |
| 关键词: 社会信任;社交推荐;异构图学习;元网络 |
|
中图分类号: TP391
文献标识码: A
|
| 基金项目: 教育部产学合作协同育人项目(220903242265640) |
|
| Application of Implicit Trust and Influencein Social Recommendation Basedon Heterogeneous Graph Convolutional Networks |
|
WANG Xiyuan, SENG Dewen
|
( School of Computer Science, Hangzhou Dianzi University, Hangzhou 310018, China)
211050083@hdu.edu.cn; sengdw@hdu.edu.cn
|
| Abstract: To address the issue of how to incorporate users' social networks into Graph Convolutional Networks (GCAN) and effectively realize heterogeneous relationship learning, this paper proposes a novel heterogeneous GCN framework that incorporates implicit trust and influence (HGCNTI). This framework constructs trust and influence subgraphs based on a user-user bipartite graph, fully leveraging the implicit relationships between users to enhance user-item representation. Additionally, a multi-view meta-network is designed to extract personalized information from different users or items, achieving adaptive enhancement of personalized knowledge transfer. Experimental results show that HGCNTI performs exceptionally well on the Ciao and Epinions datasets. Compared to various state-of-the-art baselines, it improves the recall@5 by 22.6%, recall@10 by 19.7%, and NDCG@10 by 19% on the Ciao dataset; on the Epinions dataset, it improves NDCG@5 by 2.9% and precision@10 by 4.5%. |
| Keywords: social trust; social recommendation; heterogeneous graph learning; meta-network |
