基于超图的自监督推荐算法  

Self-supervised recommender algorithm based on hypergraph

作　　者：贾小暾 温明 杨晓龙 陈宝涛 李爱荣 任媛媛 JIA Xiao-tun;WEN Ming;YANG Xiao-long;CHEN Bao-tao;LI Ai-rong;REN Yuan-yuan(College of Computer Science and Technology,Xinjiang Normal University,Urumqi 830054,China;Software Division,Xinjiang Electronic Research Institute,Urumqi 830013,China;Professional and Technical Department,Department of Human Resources and Social Security of Xinjiang,Urumqi 830011,China)

机构地区：[1]新疆师范大学计算机科学技术学院,新疆乌鲁木齐830054 [2]新疆电子研究所股份有限公司软件事业部,新疆乌鲁木齐830013 [3]新疆维吾尔自治区人力资源和社会保障厅专技处,新疆乌鲁木齐830011

出　　处：《计算机工程与设计》2025年第3期834-840,共7页Computer Engineering and Design

基　　金：新疆维吾尔自治区重点研发计划基金项目(2022B01007-2)。

摘　　要：为改善基于图神经网络的推荐模型在实际推荐场景中面临数据出现噪声和倾斜分布时性能下降的问题,提出一种基于超图的自监督推荐算法。采用超图Transformer捕捉用户与物品之间的全局关系,引入自监督学习以增强数据,提高模型的鲁棒性。在实际数据集上的训练结果表明,模型在提升推荐效果方面表现优异,特别是在解决数据稀疏性和噪声问题上表现出较强的能力。通过消融实验进一步验证了这些发现,展现了该算法在现代推荐系统中的应用潜力。To address the performance degradation of recommendation models based on graph neural networks in real-world scenarios where data exhibit noise and skewed distributions,a self-supervised recommendation algorithm based on hypergraphs was proposed.A hypergraph Transformer was employed to capture the global relationships between users and items,the self-supervised learning was introduced to enhance the data and improve the model robustness.Training results on real-world datasets demonstrate the outstanding performances of the model in enhancing recommendation effectiveness,particularly in addres-sing issues related to data sparsity and noise.Further the validation of these findings was provided through ablation experiments,highlighting its potential application in modern recommendation systems.

关 键 词：噪声数据 推荐算法 超图 全局关系 自监督学习 交互图 数据增强 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]