对比学习增强的多行为超图神经网络推荐模型

打开文本图片集

关键词：推荐系统；多行为推荐；图神经网络；超图；对比学习；自监督学习中图分类号：TP391 文献标志码：A 文章编号：1001-3695（2025）08-008-2304-08doi：10.19734/j.issn.1001-3695.2024.12.0528

Multi-behavior hypergraph neural network model enhanced with contrastive learning

WangGuang，Li Jiaxin† （CollegeofSoftware，Liaoning Technical University，Huludao Liaoning 1251O5，China）

Abstract：Multi-behaviorrecommendation（MBR）systemsareincreasinglyimportantininternetplatformsbutface twocriticallimitations：a）failuretocharacterizeusers’complexpreferencesunderdiversebehaviors，b）dificultymodeling interbehaviorrelationships.This studyproposedamulti-behaviorhypergraph neural networkmodel enhanced withcontrastielearning（MBHCL）toaddress these issues.The methodconstructed user-item hypergraphsformulti-behavior interactions，capturingusers’multi-dimensionalpreferences.Itdesigned threecontrastivetaskstointegratesingle-behaviorrepresentations throughcommonality-diferencemodeling，obtainingoptimizedembeddingstoalleviatecold-startanddatasparsityproblems. Experimentsonfourreal-worlddatasets（Tmall，BeiBei，Kuairand，Yelp）demonstrateMBHCL’sefectiveness.Themodel achieved minimum 4.8% HIT and NDCG improvements on Tmall and BeiBei datasets，and 3.6% enhancements on Kuairand andYelpdatasets.Ablationtestsverifiedallcomponents’contributions，withcold-startrecommendationsshowingsignificant performance gains.

Keywords：recommendation system；multi-behaviorrecommendation；graphneuralnetwork；hypergraph；contrastive learning；self-supervised learning

0引言

在信息爆炸的时代背景下，推荐系统作为应对信息过载、提升用户体验的关键工具，其发展受到了广泛关注。（剩余17817字）