《计算机应用》唯一官方网站 ›› 2025, Vol. 45 ›› Issue (4): 1061-1068.DOI: 10.11772/j.issn.1001-9081.2024030393
收稿日期:2024-04-08
									
				
											修回日期:2024-05-26
									
				
											接受日期:2024-05-29
									
				
											发布日期:2024-08-15
									
				
											出版日期:2025-04-10
									
				
			通讯作者:
					温鑫瑜
							作者简介:党伟超(1974—),男,山西运城人,副教授,博士,CCF会员,主要研究方向:智能计算、软件可靠性基金资助:
        
                                                                                                                            Weichao DANG, Xinyu WEN( ), Gaimei GAO, Chunxia LIU
), Gaimei GAO, Chunxia LIU
			  
			
			
			
                
        
    
Received:2024-04-08
									
				
											Revised:2024-05-26
									
				
											Accepted:2024-05-29
									
				
											Online:2024-08-15
									
				
											Published:2025-04-10
									
			Contact:
					Xinyu WEN   
							About author:DANG Weichao, born in 1974, Ph. D., associate professor. His research interests include intelligent computing, software reliability.Supported by:摘要:
针对图协同过滤推荐方法存在的单一视图局限性和数据稀疏性问题,提出一种基于多视图多尺度对比学习的图协同过滤(MVMSCL)模型。首先,根据用户-项目交互构建初始交互图,并考虑用户-项目中存在的多种潜在意图,以构建多意图分解视图;其次,利用高阶关系改进邻接矩阵,以构建协同邻居视图;再次,去除不重要的噪声交互,以构建自适应增强的初始交互图和多意图分解视图;最后,引入局部、跨层和全局3种尺度的对比学习范式生成自监督信号,从而提高推荐性能。在Gowalla、Amazon-book和Tmall 3个公共数据集上的实验结果表明,MVMSCL的推荐性能均优于对比模型。与最优基线模型DCCF(Disentangled Contrastive Collaborative Filtering framework)相比,MVMSCL的召回率Recall@20分别提升了5.7%、14.5%和10.0%,归一化折损累计增益NDCG@20分别提升了4.6%、17.9%和11.5%。
中图分类号:
党伟超, 温鑫瑜, 高改梅, 刘春霞. 基于多视图多尺度对比学习的图协同过滤[J]. 计算机应用, 2025, 45(4): 1061-1068.
Weichao DANG, Xinyu WEN, Gaimei GAO, Chunxia LIU. Multi-view and multi-scale contrastive learning for graph collaborative filtering[J]. Journal of Computer Applications, 2025, 45(4): 1061-1068.
| 数据集 | 用户数 | 项目数 | 交互数 | 稀疏度/% | 
|---|---|---|---|---|
| Gowalla | 50 821 | 57 440 | 1 172 425 | 0.040 | 
| Amazon-book | 78 578 | 77 801 | 2 240 156 | 0.037 | 
| Tmall | 47 939 | 41 390 | 2 357 450 | 0.119 | 
表1 实验数据集的统计信息
Tab. 1 Statistical information of experimental datasets
| 数据集 | 用户数 | 项目数 | 交互数 | 稀疏度/% | 
|---|---|---|---|---|
| Gowalla | 50 821 | 57 440 | 1 172 425 | 0.040 | 
| Amazon-book | 78 578 | 77 801 | 2 240 156 | 0.037 | 
| Tmall | 47 939 | 41 390 | 2 357 450 | 0.119 | 
| 模型 | Gowalla | Amazon-book | Tmall | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| R@20 | R@40 | N@20 | N@40 | R@20 | R@40 | N@20 | N@40 | R@20 | R@40 | N@20 | N@40 | |
| NCF | 0.124 7 | 0.191 0 | 0.065 9 | 0.083 2 | 0.046 8 | 0.077 1 | 0.033 6 | 0.043 8 | 0.038 3 | 0.064 7 | 0.025 2 | 0.034 4 | 
| AutoR | 0.140 9 | 0.214 2 | 0.071 6 | 0.090 5 | 0.054 6 | 0.091 4 | 0.035 4 | 0.048 2 | 0.033 6 | 0.061 1 | 0.020 3 | 0.029 5 | 
| NGCF | 0.141 3 | 0.207 2 | 0.081 3 | 0.098 7 | 0.053 2 | 0.086 6 | 0.038 8 | 0.050 1 | 0.042 0 | 0.075 1 | 0.025 0 | 0.036 5 | 
| LightGCN | 0.179 9 | 0.257 7 | 0.105 3 | 0.125 5 | 0.073 2 | 0.114 8 | 0.054 4 | 0.068 1 | 0.055 5 | 0.089 5 | 0.038 1 | 0.049 9 | 
| DGCF | 0.178 4 | 0.251 5 | 0.106 9 | 0.125 9 | 0.068 8 | 0.107 3 | 0.051 3 | 0.064 0 | 0.054 4 | 0.086 7 | 0.037 2 | 0.048 4 | 
| DGCL | 0.179 3 | 0.248 3 | 0.106 7 | 0.124 7 | 0.067 7 | 0.105 7 | 0.050 6 | 0.063 1 | 0.052 6 | 0.084 5 | 0.035 9 | 0.046 9 | 
| SGL-ED | 0.180 9 | 0.255 9 | 0.106 7 | 0.126 2 | 0.077 4 | 0.120 4 | 0.057 8 | 0.071 9 | 0.057 4 | 0.091 9 | 0.039 3 | 0.051 3 | 
| SGL-ND | 0.181 4 | 0.258 9 | 0.106 5 | 0.126 7 | 0.072 2 | 0.112 1 | 0.054 2 | 0.067 4 | 0.055 3 | 0.088 5 | 0.037 9 | 0.049 4 | 
| HCCF | 0.181 8 | 0.260 1 | 0.106 1 | 0.126 5 | 0.082 4 | 0.128 2 | 0.062 5 | 0.077 6 | 0.062 3 | 0.098 6 | 0.042 5 | 0.055 2 | 
| LightGCL | 0.182 5 | 0.260 1 | 0.107 7 | 0.128 0 | 0.083 6 | 0.128 0 | 0.064 3 | 0.079 0 | 0.063 2 | 0.097 1 | 0.044 4 | 0.056 2 | 
| DCCF | ||||||||||||
| MVMSCL | 0.198 3 | 0.274 2 | 0.117 5 | 0.137 4 | 0.101 8 | 0.148 9 | 0.080 2 | 0.099 5 | 0.073 5 | 0.113 2 | 0.052 3 | 0.066 0 | 
表2 不同模型在公共数据集上的性能对比
Tab. 2 Performance comparison of different models on public datasets
| 模型 | Gowalla | Amazon-book | Tmall | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| R@20 | R@40 | N@20 | N@40 | R@20 | R@40 | N@20 | N@40 | R@20 | R@40 | N@20 | N@40 | |
| NCF | 0.124 7 | 0.191 0 | 0.065 9 | 0.083 2 | 0.046 8 | 0.077 1 | 0.033 6 | 0.043 8 | 0.038 3 | 0.064 7 | 0.025 2 | 0.034 4 | 
| AutoR | 0.140 9 | 0.214 2 | 0.071 6 | 0.090 5 | 0.054 6 | 0.091 4 | 0.035 4 | 0.048 2 | 0.033 6 | 0.061 1 | 0.020 3 | 0.029 5 | 
| NGCF | 0.141 3 | 0.207 2 | 0.081 3 | 0.098 7 | 0.053 2 | 0.086 6 | 0.038 8 | 0.050 1 | 0.042 0 | 0.075 1 | 0.025 0 | 0.036 5 | 
| LightGCN | 0.179 9 | 0.257 7 | 0.105 3 | 0.125 5 | 0.073 2 | 0.114 8 | 0.054 4 | 0.068 1 | 0.055 5 | 0.089 5 | 0.038 1 | 0.049 9 | 
| DGCF | 0.178 4 | 0.251 5 | 0.106 9 | 0.125 9 | 0.068 8 | 0.107 3 | 0.051 3 | 0.064 0 | 0.054 4 | 0.086 7 | 0.037 2 | 0.048 4 | 
| DGCL | 0.179 3 | 0.248 3 | 0.106 7 | 0.124 7 | 0.067 7 | 0.105 7 | 0.050 6 | 0.063 1 | 0.052 6 | 0.084 5 | 0.035 9 | 0.046 9 | 
| SGL-ED | 0.180 9 | 0.255 9 | 0.106 7 | 0.126 2 | 0.077 4 | 0.120 4 | 0.057 8 | 0.071 9 | 0.057 4 | 0.091 9 | 0.039 3 | 0.051 3 | 
| SGL-ND | 0.181 4 | 0.258 9 | 0.106 5 | 0.126 7 | 0.072 2 | 0.112 1 | 0.054 2 | 0.067 4 | 0.055 3 | 0.088 5 | 0.037 9 | 0.049 4 | 
| HCCF | 0.181 8 | 0.260 1 | 0.106 1 | 0.126 5 | 0.082 4 | 0.128 2 | 0.062 5 | 0.077 6 | 0.062 3 | 0.098 6 | 0.042 5 | 0.055 2 | 
| LightGCL | 0.182 5 | 0.260 1 | 0.107 7 | 0.128 0 | 0.083 6 | 0.128 0 | 0.064 3 | 0.079 0 | 0.063 2 | 0.097 1 | 0.044 4 | 0.056 2 | 
| DCCF | ||||||||||||
| MVMSCL | 0.198 3 | 0.274 2 | 0.117 5 | 0.137 4 | 0.101 8 | 0.148 9 | 0.080 2 | 0.099 5 | 0.073 5 | 0.113 2 | 0.052 3 | 0.066 0 | 
| 模型 | R@20 | R@40 | N@20 | N@40 | 
|---|---|---|---|---|
| MVMSCL- W | 0.184 2 | 0.258 6 | 0.108 2 | 0.127 7 | 
| MVMSCL- | 0.190 7 | 0.268 0 | 0.112 4 | 0.132 6 | 
| MVMSCL- | 0.190 9 | 0.268 1 | 0.112 5 | 0.132 6 | 
| MVMSCL- T | 0.187 7 | 0.264 4 | 0.112 1 | 0.132 9 | 
| MVMSCL-L | 0.189 5 | 0.266 0 | 0.112 1 | 0.132 1 | 
| MVMSCL-G | 0.195 8 | 0.272 7 | 0.116 1 | 0.136 3 | 
| MVMSCL-C | 0.198 0 | 0.274 0 | 0.117 0 | 0.137 1 | 
| MVMSCL | 0.198 3 | 0.274 2 | 0.117 5 | 0.137 4 | 
表3 MVMSCL模型消融实验对比结果
Tab. 3 Comparison results of MVMSCL model’s ablation experiments
| 模型 | R@20 | R@40 | N@20 | N@40 | 
|---|---|---|---|---|
| MVMSCL- W | 0.184 2 | 0.258 6 | 0.108 2 | 0.127 7 | 
| MVMSCL- | 0.190 7 | 0.268 0 | 0.112 4 | 0.132 6 | 
| MVMSCL- | 0.190 9 | 0.268 1 | 0.112 5 | 0.132 6 | 
| MVMSCL- T | 0.187 7 | 0.264 4 | 0.112 1 | 0.132 9 | 
| MVMSCL-L | 0.189 5 | 0.266 0 | 0.112 1 | 0.132 1 | 
| MVMSCL-G | 0.195 8 | 0.272 7 | 0.116 1 | 0.136 3 | 
| MVMSCL-C | 0.198 0 | 0.274 0 | 0.117 0 | 0.137 1 | 
| MVMSCL | 0.198 3 | 0.274 2 | 0.117 5 | 0.137 4 | 
| 模型 | R@20 | R@40 | N@20 | N@40 | 
|---|---|---|---|---|
| 0.136 6 | 0.198 7 | 0.081 4 | 0.097 6 | |
| 0.172 3 | 0.245 5 | 0.101 7 | 0.121 1 | |
| 0.171 9 | 0.245 1 | 0.101 4 | 0.120 5 | |
| 0.166 3 | 0.241 1 | 0.097 2 | 0.116 6 | |
| 0.184 2 | 0.258 6 | 0.108 2 | 0.127 2 | 
表4 多视图融合策略实验结果
Tab. 4 Experimental results of multi-view fusion strategies
| 模型 | R@20 | R@40 | N@20 | N@40 | 
|---|---|---|---|---|
| 0.136 6 | 0.198 7 | 0.081 4 | 0.097 6 | |
| 0.172 3 | 0.245 5 | 0.101 7 | 0.121 1 | |
| 0.171 9 | 0.245 1 | 0.101 4 | 0.120 5 | |
| 0.166 3 | 0.241 1 | 0.097 2 | 0.116 6 | |
| 0.184 2 | 0.258 6 | 0.108 2 | 0.127 2 | 
| 模型 | R@20 | R@40 | N@20 | N@40 | 
|---|---|---|---|---|
| AllLocal+Layer | 0.180 9 | 0.106 8 | 0.256 2 | 0.126 5 | 
| AllGlobal+Layer | 0.192 1 | 0.114 6 | 0.269 5 | 0.134 8 | 
| Local+Global+Layer | 0.198 3 | 0.117 5 | 0.274 2 | 0.137 4 | 
表5 多尺度对比策略
Tab. 5 Multi-scale comparative strategies
| 模型 | R@20 | R@40 | N@20 | N@40 | 
|---|---|---|---|---|
| AllLocal+Layer | 0.180 9 | 0.106 8 | 0.256 2 | 0.126 5 | 
| AllGlobal+Layer | 0.192 1 | 0.114 6 | 0.269 5 | 0.134 8 | 
| Local+Global+Layer | 0.198 3 | 0.117 5 | 0.274 2 | 0.137 4 | 
| 1 | GAO C, ZHENG Y, LI N, et al. A survey of graph neural networks for recommender systems: challenges, methods, and directions[J]. ACM Transactions on Recommender Systems, 2023, 1(1): No.3. | 
| 2 | PARK S, YOON M, LEE J W, et al. Toward a better understanding of loss functions for collaborative filtering [C]// Proceedings of the 32nd ACM International Conference on Information and Knowledge Management. New York: ACM, 2023: 2034-2043. | 
| 3 | WANG X, HE X, WANG M, et al. Neural graph collaborative filtering [C]// Proceedings of the 42nd International ACM SIGIR Conference on Research and Development in Information Retrieval. New York: ACM, 2019: 165-174. | 
| 4 | HE X, DENG K, WANG X, et al. LightGCN: simplifying and powering graph convolution network for recommendation [C]// Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval. New York: ACM, 2020: 639-648. | 
| 5 | MAO K, ZHU J, XIAO X, et al. UltraGCN: ultra simplification of graph convolutional networks for recommendation [C]// Proceedings of the 30th ACM International Conference on Information and Knowledge Management. New York: ACM, 2021: 1253-1262. | 
| 6 | LINSKER R. Self-organization in a perceptual network [J]. Computer, 1988, 21(3):105-117. | 
| 7 | 杨兴耀,陈羽,于炯,等. 结合自我特征和对比学习的推荐模型[J]. 计算机应用, 2024, 44(9):2704-2710. | 
| YANG X Y, CHEN Y, YU J, et al. Recommendation model combining self-features and contrastive learning [J]. Journal of Computer Applications, 2024, 44(9):2704-2710. | |
| 8 | WU J, WANG X, FENG F, et al. Self-supervised graph learning for recommendation [C]// Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval. New York: ACM, 2021: 726-735. | 
| 9 | LIN Z, TIAN C, HOU Y, et al. Improving graph collaborative filtering with neighborhood-enriched contrastive learning [C]// Proceedings of the ACM Web Conference 2022. New York: ACM, 2022: 2320-2329. | 
| 10 | XIA L, HUANG C, XU Y, et al. Hypergraph contrastive collaborative filtering [C]// Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval. New York: ACM, 2022: 70-79. | 
| 11 | WANG W, FENG F, HE X, et al. Deconfounded recommendation for alleviating bias amplification [C]// Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. New York: ACM, 2021: 1717-1725. | 
| 12 | ABRAMOVICH S, PERSSON L E. Some new estimates of the ‘Jensen gap’ [J]. Journal of Inequalities and Applications, 2016, 2016: No.39. | 
| 13 | GAO X, SITHARAM M, ROITBERG A E. Bounds on the Jensen gap, and implications for mean-concentrated distributions [J]. Information Processing and Management, 2019, 16(2): No.14. | 
| 14 | HAMERS L, HEMERYCK Y, HERWEYERS G, et al. Similarity measures in scientometric research: the Jaccard index versus Salton’s cosine formula [J]. Information Processing and Management, 1989, 25(3): 315-318. | 
| 15 | TIAN C, XIE Y, LI Y, et al. Learning to denoise unreliable interactions for graph collaborative filtering [C]// Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval. New York: ACM, 2022: 122-132. | 
| 16 | REN X, XIA L, ZHAO J, et al. Disentangled contrastive collaborative filtering [C]// Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval. New York: ACM, 2023:1137-1146. | 
| 17 | CAI X, HUANG C, XIA L, et al. LightGCL: simple yet effective graph contrastive learning for recommendation [EB/OL]. [2024-05-23]. . | 
| 18 | HE X, LIAO L, ZHANG H, et al. Neural collaborative filtering[C]// Proceedings of the 26th International Conference on World Wide Web. Republic and Canton of Geneva: International World Wide Web Conferences Steering Committee, 2017: 173-182. | 
| 19 | SEDHAIN S, MENON A K, SANNER S, et al. AutoRec: autoencoders meet collaborative filtering [C]// Proceedings of the 24th International Conference on World Wide Web. New York: ACM, 2015: 111-112. | 
| 20 | WANG X, JIN H, ZHANG A, et al. Disentangled graph collaborative filtering [C]// Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval. New York: ACM, 2020: 1001-1010. | 
| 21 | LI H, WANG X, ZHANG Z, et al. Disentangled contrastive learning on graphs [C]// Proceedings of the 35th International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2021: 21872-21884. | 
| [1] | 游兰, 张雨昂, 刘源, 陈智军, 王伟, 曾星, 何张玮. 基于协作贡献网络的开源项目开发者推荐[J]. 《计算机应用》唯一官方网站, 2025, 45(4): 1213-1222. | 
| [2] | 王聪, 史艳翠. 基于多视角学习的图神经网络群组推荐模型[J]. 《计算机应用》唯一官方网站, 2025, 45(4): 1205-1212. | 
| [3] | 杨光局, 罗天健, 王开军, 杨思琪. 多分支多视图的时间序列上下文对比表征学习方法[J]. 《计算机应用》唯一官方网站, 2025, 45(4): 1042-1052. | 
| [4] | 田仁杰, 景明利, 焦龙, 王飞. 基于混合负采样的图对比学习推荐算法[J]. 《计算机应用》唯一官方网站, 2025, 45(4): 1053-1060. | 
| [5] | 王元龙, 刘亭华, 张虎. 基于跨模态对比学习的常识问答模型[J]. 《计算机应用》唯一官方网站, 2025, 45(3): 732-738. | 
| [6] | 陈维, 施昌勇, 马传香. 基于多模态数据融合的农作物病害识别方法[J]. 《计算机应用》唯一官方网站, 2025, 45(3): 840-848. | 
| [7] | 蔡启健, 谭伟. 语义图增强的多模态推荐算法[J]. 《计算机应用》唯一官方网站, 2025, 45(2): 421-427. | 
| [8] | 杨晟, 李岩. 面向目标检测的对比知识蒸馏方法[J]. 《计算机应用》唯一官方网站, 2025, 45(2): 354-361. | 
| [9] | 严雪文, 黄章进. 基于对比学习的小样本图像分类方法[J]. 《计算机应用》唯一官方网站, 2025, 45(2): 383-391. | 
| [10] | 马汉达, 吴亚东. 多域时空层次图神经网络的空气质量预测[J]. 《计算机应用》唯一官方网站, 2025, 45(2): 444-452. | 
| [11] | 区卓越, 邓秀勤, 陈磊. 基于加权锚点的自适应多视图互补聚类算法[J]. 《计算机应用》唯一官方网站, 2025, 45(1): 115-126. | 
| [12] | 赵文博, 马紫彤, 杨哲. 基于有向超图自适应卷积的链接预测模型[J]. 《计算机应用》唯一官方网站, 2025, 45(1): 15-23. | 
| [13] | 余肖生, 王智鑫. 基于多层次图对比学习的序列推荐模型[J]. 《计算机应用》唯一官方网站, 2025, 45(1): 106-114. | 
| [14] | 程子栋, 李鹏, 朱枫. 物联网威胁情报知识图谱中潜在关系的挖掘[J]. 《计算机应用》唯一官方网站, 2025, 45(1): 24-31. | 
| [15] | 杜郁, 朱焱. 构建预训练动态图神经网络预测学术合作行为消失[J]. 《计算机应用》唯一官方网站, 2024, 44(9): 2726-2731. | 
| 阅读次数 | ||||||
| 全文 |  | |||||
| 摘要 |  | |||||