融合用户属性的多层次对比学习知识感知推荐方法

打开文本图片集

关键词：推荐系统；知识图谱；对比学习；图神经网络

中图分类号：TP391 文献标志码：A 文章编号：1001-3695（2026）01-009-0076-07

doi：10.19734/j.issn.1001-3695.2025.06.0207

Multi-level contrastive learning for knowledge-aware recommendation with user attribute fusion

Cao Chunping†，Wen Xinyu （SchoolofOicalElerical&uerEgering，Uesityfghaiforiece&hologioina）

Abstract：Existing knowledgegraph-basedrecommendation modelsoftenfail tofullyexploituseratributesandcapture highorderinformation，andtheyeasilyintroduceredundantsignalsthatreducerecommendationperformance.Toaddresstheseissues，this studyproposedacontrastivelearningrecommendationmodel，caledMACRec（meta-guidedatribute-awarecontrastiverecommendationmodel），which integrateduserattibutes.Themodelconstructedanatributeviewto incorporateusersidefeatures，thereby enhancing thediversityand accuracy ofnoderepresentations.It designeda meta-path-guided domain constructionstrategytofilterhigh-orderneighborsandstrengthenstructural modeling.Itfurtherintroducedsame-orderand cross-ordercontrastivemechanisms tobalancecollaborativesignalswithknowledgegraphsignals.ExperimentsontheMovieLens1M andBook-Crossng datasets show thatthis modeloutperforms mainstream methods on multiple metrics，whichdemonstrates its effectiveness in improving both recommendation accuracy and generalization ability.

Key words： recommender system； knowledge graph；contrastive learning；graph neural network（GNN

0 引言

面对信息过载的挑战，推荐系统凭借其在海量信息中挖掘用户感兴趣内容、提升获取信息效率的能力，成为学术界和工业界研究的热点。（剩余19480字）