1.Network Security Department, Shanxi Police College, Taiyuan 030401, China
2.Key Laboratory of Computational Intelligence and Chinese Information Processing of Ministry of Education, Shanxi University, Taiyuan 030006, China
Show less
文章历史+
Received
Accepted
Published
2024-09-02
2025-03-08
Issue Date
2025-10-09
PDF (3686K)
摘要
在推荐系统中,由于用户评分和交互频次存在长尾分布特性,导致长尾物品特征提取困难,现有的方法或者过多关注尾部物品,忽略其与头部物品的联系,或者忽略社交网络对用户偏好的影响,从而影响了推荐效果。因此,本文提出一种融合社会影响扩散的长尾推荐模型LoSidi(Long-tail Recommendation method Incorporating Social Influence Diffusion)。首先对于每个用户,通过对目标用户各层社交邻居的采样聚合,并结合其已交互的热门物品生成用户兴趣嵌入。其次通过计算长尾物品与用户已交互头部物品间的相似性挖掘长尾物品的潜在特征。最终,LoSidi模型构建用户与长尾物品间的联系,对长尾物品进行评分预测和推荐。在常用数据集上的实验结果表明,本文提出的模型可显著提高用户推荐列表的新颖性和多样性。
Abstract
In recommendation systems, the long-tail distribution of user ratings and interaction frequencies poses challenges for extracting the features of long-tail items. Existing methods either overly focus on tail items while neglecting their connections with head items or disregard the influence of social networks on user preferences, thereby impacting recommendation performance. To address these issues, this paper proposes a novel long-tail recommendation model called LoSidi (Long-tail Recommendation Method Incorporating Social Influence Diffusion). Firstly, for each user, the model aggregates samples of social neighbors at various layers and integrates these with the popular items the user has interacted with to generate user interest embeddings. Secondly, the potential features of long-tail items are mined by calculating the similarity between long-tail items and the head items the user has interacted with. Finally, the LoSidi model establishes links between users and long-tail items, predicting scores and generating recommendations for these items. Experimental results on widely-used datasets demonstrate that the proposed model significantly improves the novelty and diversity of users' recommendation lists.
社交图谱蕴含的异构用户关联网络为推荐系统提供了超越传统用户-商品交互矩阵的潜在兴趣表征源,这一发现驱动着融合社交网络的推荐模型成为近年来的研究热点。具体来说,SoRec[10](Social Recommendation Using Probabilistic Matrix Factorization)率先将社交拓扑网络整合至概率矩阵分解框架,通过异构数据联合建模有效缓解用户-商品交互矩阵的稀疏性挑战,显著提升评分预测精度。该工作启发了SocialRSTE[11](Recommendation with Social Trust Ensemble),该框架构建信任感知的协同过滤机制,创新性地建立用户本体偏好与社交信任网络成员偏好的贝叶斯耦合关系。研究前沿进一步延伸至正则化技术层面,SoReg[12] (Recommender Systems with Social Regularization)通过设计基于图拉普拉斯约束的社交正则化项,在矩阵分解目标函数中编码用户社交近邻的隐式相似性先验。SocialMF[13](Matrix Factorization Based Model for Recommendation in Social Rating Networks)模型在SoRec基础上进行改进,通过引入加权平均机制学习用户特征。TrustMF[14](Social Collaborative Filtering by Trust)进一步从双向信任关系(信任与被信任)的角度构建用户表征,从而提升对未知项目评分的预测精度。LOCABAL[15](Exploiting Local and Global Social Context for Recommendation)提出分层社交上下文建模框架,通过微观用户邻域特征提取与宏观社群结构感知的双流架构,创新性地解耦局部-全局社交影响力传播路径。TrustSVD[16](A Trust-based Matrix Factorization Technique)模型在SVD++[17]框架上进行扩展,创新性地融合了显式与隐式双重社交信息[18],不仅整合用户的显式评分和直接社交关系,还引入用户隐式行为数据和间接社交互动,从而构建了更全面的推荐模型。这些社交推荐方法旨在处理一般的推荐问题,追求推荐精度的提高,但没有考虑对长尾商品的推荐。
社交影响力的传播特性推动了图神经网络(Graph Neural Network, GNN)在推荐领域的创新应用。例如,研究者通过动态图表示方法构建社交信息传播路径,结合多层级邻居特征聚合技术,更精准地捕捉用户行为背后的社交驱动因素。典型的方法有DiffNet[19](A neural Influence Diffusion Model for Social Recommendation),提出基于图采样技术GraphSAGE[20](Inductive Representation Learning on Large Graphs)的社交影响力扩散模型,通过融合用户社交圈特征与历史交互行为数据生成动态用户表征。GraphRec[21](Graph Neural Networks for Social Recommendation)则构建异构信息融合框架,从社交关系网和用户-商品交互图中联合学习嵌入表示,其中注意力网络可量化不同社交邻居的贡献权重。这些方法将用户之间的关系与用户项目的关系转化为图结构,利用GNN获取用户节点和项目节点的潜在信息,为推荐系统的研究提供了新的方向。但当前研究在社交网络与长尾推荐的融合方向仍存在双重局限:其一,现有方法多聚焦于直接社交关联(如单跳邻居建模),却忽视社交影响力的多跳传播机制;其二,用户偏好建模缺乏动态适应性,尤其在长尾场景下难以精准捕捉跨层级的兴趣演化路径。
为验证本文方法对长尾推荐的有效性,选择常用对比方法SoRec[10]、SoReg[12]、TrustSVD[18]、RIITD(Recommendation Model Combining Implicit Influence of Trust with Trust Degree)[22]与本文方法进行比较。这些方法都利用了社交信息处理推荐问题,其特点如下。
BRYNJOLFSSONE, HUY, SIMESTERD. Goodbye Pareto Principle, Hello Long Tail: The Effect of Search Costs on the Concentration of Product Sales[J]. Management Science, 2011, 57(8): 1373-1386. DOI: 10.1287/mnsc.1110.1371 .
VALCARCED, PARAPARJ, BARREIROÁ. Item-based Relevance Modelling of Recommendations for Getting Rid of Long Tail Products[J]. Knowl Based Syst, 2016, 103: 41-51. DOI: 10.1016/j.knosys.2016.03.021 .
[6]
OESTREICHER-SINGERG, SUNDARARAJANA. Recommendation Networks and the Long Tail of Electronic Commerce[J]. MIS Quarterly, 2012, 36(1):65-83. DOI: 10.2139/ssrn.1324064 .
[7]
HEH B, GARCIAE A. Learning from Imbalanced Data[J]. IEEE Trans Knowl Data Eng, 2009, 21(9): 1263-1284. DOI: 10.1109/TKDE.2008.239 .
[8]
CUIY, JIAM L, LINT Y, et al. Class-balanced Loss Based on Effective Number of Samples[C]//2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE, 2019: 9260-9269. DOI: 10.1109/CVPR.2019.00949 .
[9]
ZHANGY, CHENGD Z, YAOT S, et al. A Model of Two Tales: Dual Transfer Learning Framework for Improved Long-tail Item Recommendation[C]//Proceedings of the Web Conference 2021. New York: ACM, 2021: 2220-2231. DOI: 10.1145/3442381.3450086 .
[10]
YINH Z, CUIB, LIJ, et al. Challenging the Long Tail Recommendation[J]. Proc VLDB Endow, 2012, 5(9): 896-907. DOI: 10.14778/2311906.2311916 .
[11]
MAH, KINGI, LYUM R. Learning to Recommend with Social Trust Ensemble[C]//Proceedings of the 32nd International ACM SIGIR Conference on Research and Development in Information Retrieval. New York: ACM, 2009: 203-210. DOI: 10.1145/1571941.1571978 .
[12]
MAH, KINGI, LYUM R. Learning to Recommend with Social Trust Ensemble[C]//Proceedings of the 32nd International ACM SIGIR Conference on Research and Development in Information Retrieval. New York, USA: Association for Computing Machinery, 2009: 203-210. DOI: 10.1145/1571941.1571978 .
[13]
MAH, ZHOUD Y, LIUC, et al. Recommender Systems with Social Regularization[C]//Proceedings of the Fourth ACM International Conference on Web Search and Data Mining. New York: ACM, 2011: 287-296. DOI: 10.1145/1935826.1935877 .
[14]
JAMALIM, ESTERM. A Matrix Factorization Technique with Trust Propagation for Recommendation in Social Networks[C]//Proceedings of the Fourth ACM Conference on Recommender Systems. New York: ACM, 2010: 135-142. DOI: 10.1145/1864708.1864736 .
[15]
YANGB, LEIY, LIUJ M, et al. Social Collaborative Filtering by Trust[J]. IEEE Trans Pattern Anal Mach Intell, 2017, 39(8): 1633-1647. DOI: 10.1109/TPAMI.2016.2605085 .
[16]
TANGJ L, HUX, GAOH J, et al. Exploiting Local and Global Social Context for Recommendation[C]// Proceedings of the Twenty-Third International Joint Conference on Artificial Intelligence. Beijing: AAAI Press, 2013: 2712-2718. DOI: 10.5555/2540128.2540519 .
[17]
GUOG B, ZHANGJ. TrustSVD: Collaborative Filtering with Both the Explicit and Implicit Influence of User Trust and of Item Ratings[C]//Proceedings of the Twenty-Ninth AAAI Conference on Artificial Intelligence. Menlo Park: Press, 2015, 29(1): 123-129. DOI: 10.1609/aaai.v29i1.9153 .
[18]
KORENY. Factorization Meets the Neighborhood: A Multifaceted Collaborative Filtering Model[C]//Proceedings of the 14th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. New York: ACM, 2008: 426-434. DOI: 10.1145/1401890.1401944 .
[19]
YAOW L, HEJ, HUANGG Y, et al. Modeling Dual Role Preferences for Trust-Aware Recommendation[C]//Proceedings of the 37th International ACM SIGIR Conference on Research & Development in Information Retrieval. New York, USA: Association for Computing Machinery, 2014: 975-978. DOI: 10.1145/2600428.2609488 .
[20]
WUL, SUNP, FUY, et al. A Neural Influence Diffusion Model for Social Recommendation[C]//Proceedings of the 42nd International ACM SIGIR Conference on Research and Development in Information Retrieval. Paris, France: ACM, 2019: 235-244. DOI: 10.1145/3331184.3331214 .
[21]
HAMILTONW, YINGZ, LESKOVECJ. Inductive Representation Learning on Large Graphs[C]//Advances in Neural Information Processing Systems. Long Beach, California, USA: Curran Associates, Inc., 2017: 1024-1034. DOI: 10.48550/arXiv.1706.02216 .
[22]
FANW, MAY, LIQ, et al. Temporal Graph Neural Networks for Social Recommendation[C]//Proceedings of the 2020 IEEE International Conference on Big Data (Big Data). Atlanta, GA, USA: IEEE, 2020: 437-446. DOI: 10.1109/BigData50022.2020.9378444 .