社交异构知识引导的多行为序列推荐方法

打开文本图片集

关键词：序列推荐；多行为；社交异构时序知识图；社交感知的高阶表示；注意力机制中图分类号：TP391 文献标志码：A 文章编号：1001-3695（2026）01-018-0153-08doi：10.19734/j. issn.1001-3695.2026.05.0133

Social heterogeneous knowledge guided multi-behavior sequential recommendation method

LiQingqing†，Chen Lei （CollegeofInformation ScienceandTechnology，GansuAgriculturalUniversity，Lanzhou73oo7o，China）

Abstract：Theexistingsequentialrecommendationmethodsoftenoverlook thesocial influenceamong usersandfail toincorporate the multi-behavior informationofuser interaction.Atthesame time，theytypicalllack acurate modelingof complex temporal dynamic features guided bysocialrelationships，includingbothusers’historical habitsanddynamicneeds.Toaddressthese chalenges，this paperdesignedan SHKM-SR，asocial heterogeneous knowledge guided multiplebehaviorsequence recommendationmethod.Specficall，themethodfirstlyintegratedtemporalinteractioninformationwithsocialrelationshipsto constructasocial heterogeneous temporal knowledge graph.Then，itencoded heterogeneous interactions withtemporalsignals andextractedhighordersocial-awarerepresentationsofnodes.Again，underthegudanceofsocialrelationships，itcaptured bothdynamiccharacteristicsandhistoricalhabitsofnodes.Itfurtherintegratedlongandshort-termpreferencesofsocial-aware preferences basedonatentionmechanismtoobtain finer graiedrepresentations.Finall，itemployedamulti-layerpeceptron to calculate itemrecommendation scoresand generate personalizedrecommendations.Theexperimentalresultson Yelp， Ciaoand Douban Book datasets show that the methodoutperforms most benchmark methods，achieving amaximum improvement of 9.6% in Hit（ω10 . The experimental results validate the effectiveness of the model in multi-behavior sequential recommendation.

Keywords：sequentialrecommendation；multi-behavior;social heterogeneous temporal knowledgegraph；high-ordersocialawarerepresentations；attention mechanism

0 引言

作为解决信息过载问题并帮助用户挖掘个性化信息的有力工具，推荐系统旨在分析用户历史交互数据，从而构建用户画像并为其推荐合适的商品，被广泛应用于电子商务、定向广告等众多现实场景中。（剩余20255字）