基于异构图的双通道交叉自适应对比学习推荐  

Recommendation Based on Graph Heterogeneous Using Dual Channel Cross-Adaptive Contrast Learning

作　　者：范伟[1,2] 周魏 文俊浩[1,2] FAN Wei;ZHOU Wei;WEN Jun-hao(School of Big Data&Software Engineering,Chongqing University,Chongqing 400044,China;Key Laboratory of Dependable Service Computing in Cyber Physical Society(Chongqing University),Ministry of Education,Chongqing 400044,China)

机构地区：[1]重庆大学大数据与软件学院,重庆400044 [2]信息物理社会可信服务计算教育部重点实验室(重庆大学),重庆400044

出　　处：《电子学报》2023年第7期1929-1938,共10页Acta Electronica Sinica

基　　金：国家自然科学基金(No.72074036,No.62072060);中国博士后科学基金(No.2020M673145);中央高校基金(No.2022CDJXY-022)。

摘　　要：通过用户多行为进行推荐任务中,各个行为通常不是独立作用的,行为之间的协同作用和依赖关系挖掘更能增强用户行为模式建模,反映用户偏好.而用户多行为关系的引入也会增加用户物品交互图与表征空间中的异质性(heterogeneity).针对上述问题,本文设计了一种基于异构图的双通道交叉自适应对比学习推荐模型MB-DCAC(Multi-Behavior Recommendation through Dual-channel Cross-Adaptive Contrast learning),创新性的从异构数据卷积过程构建对比学习方案,并基于异构连接进行表征属性增强,以提升模型挖掘用户行为模式与表达能力.实验结果表明,本文模型在Tmall、IJCAI-Context、Beibei三个数据集上,相较于基准模型在HR@10指标上分别提升了16.7%、18.3%、2.76%.且模型在挖掘多行为之间的依赖挖掘等任务上表现优异.In the task of recommendation through multiple behaviors,individual behaviors usually do not work inde⁃pendently.The mining of collaborative effects and dependencies between behaviors can better enhance user behavior mod⁃eling and reflect user preferences.The introduction of multiple user behavior relationships also increases the heterogeneity in the user-item interaction graph and representation space.To address these issues,this paper proposes a dual-channel cross-adaptive contrast learning recommendation model(MB-DCAC),based on heterogeneous graphs.To improve the model's ability to mine user behavior patterns and expressions,this model innovatively constructs a comparative learning scheme from the convolution process of heterogeneous data and enhances the representation features based on heteroge⁃neous connections.Experimental results show that compared with the baseline model,the proposed model improves the HR@10 metric by 16.7%,18.3%,and 2.76%on the Tmall,IJCAI-Context,and Beibei datasets,respectively.This model al⁃so performs well in tasks such as mining dependencies between multiple behaviors.

关 键 词：推荐系统 多行为 异构图 对比学习 行为依赖 异质性 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]