移动通信2024，Vol.48Issue(3)：152-156,5.DOI:10.3969/j.issn.1006-1010.20221117-0001
基于多视图对比学习的多行为推荐
Multi-behavior Recommendation Based on Multi-view Contrastive Learning
摘要
Abstract
In order to effectively extract semantic and structural information from user behavior graphs and fully utilize the complex dependencies between nodes,we propose a multi-behavior recommendation method(MVCL)based on multi-view contrastive learning.MVCL constructs a meta-path view and structure view to model high-order semantic information and local structural information between nodes respectively.In addition,we use self-supervised techniques and propose a cross-view contrastive learning mechanism,which enables the two views to collaborate with each other.Compared with multiple baseline models on real-world datasets,MVCL achieves the highest normalized discounted cumulative gain improvement of 24.7%.Experimental results demonstrate that MVCL outperforms other models and can learn more effective node representations.关键词
推荐系统/图注意力网络/对比学习/元路径Key words
recommendation system/graph attention network/contrastive learning/meta-path分类
信息技术与安全科学引用本文复制引用
魏静,李剑..基于多视图对比学习的多行为推荐[J].移动通信,2024,48(3):152-156,5.基金项目
国家自然科学基金(92046001) （92046001）
中央高校基础研究基金(2019XD-A02) （2019XD-A02）
