计算机应用研究2025，Vol.42Issue(11)：3265-3274,10.DOI:10.19734/j.issn.1001-3695.2025.05.0125
基于图扩散增强的对比学习推荐系统框架
Graph diffusion-augmented based contrastive learning framework for recommendation systems
摘要
Abstract
Graph neural network based recommendation systems have achieved significant process during the past few years.However,existing self-supervised contrastive learning methods construct augmented views via removing edges randomly.Conse-quently,it is difficult to distinguish edge important,which limits model performance.To address this issue,this paper proposed a novel recommendation system scheme that exploited graph diffusion-augmented contrastive learning(GDACL)to improve the recommendation performance in the data sparsity and noise scenarios.GDACL firstly preserved the information about edge weight using graph diffusion,and then enhanced the diversity and robustness of features through integrating graph convolution embedding concatenation,random feature masking,and contrastive learning.Next,it designed a gating controlling mechanism to adaptively fuse representations from the original and diffusion graphs.Experimental results show that GDACL outperforms com-peting methods across multiple real-world benchmark datasets,and exhibits stronger robustness and generalization ability.Abla-tion studies further verify the contributions of gating controlling mechanism,random feature masking and contrastive learnings between the original and diffusion graphs to performance improvement.关键词
推荐系统/对比学习/数据增强/门控机制Key words
recommendation system/contrastive learning/data augmentation/gating mechanism分类
计算机与自动化引用本文复制引用
汤家谱,李曼,曹文明,杨明明,蒋明阳..基于图扩散增强的对比学习推荐系统框架[J].计算机应用研究,2025,42(11):3265-3274,10.基金项目
国家自然科学基金资助项目(62306052,12401678) （62306052,12401678）
重庆市教育委员会科学技术研究项目(KJQN202300735,KJQN202400716) （KJQN202300735,KJQN202400716）
