基于域内和域间元路径聚合的跨域推荐方法

打开文本图片集

关键词：跨域推荐；冷启动问题；图神经网络；异构信息网络；元路径中图分类号：TP311 文献标志码：A 文章编号：1001-3695（2025）08-017-2374-09doi：10.19734/j.issn.1001-3695.2025.01.0022

Cross-domain recommendation method based on aggregation of intra-domain and inter-domain meta-paths

Xu Jia 1，2 ，Wang Xin³，Wang Yanqiu4†，Wu Haiwei³，Lyu Pin1 （1.CyberspaceuteofdadlogouUesityaga;ateKybotoyse curityDefesedsof Inforationii；ofceleit Guangdong 518055，China）

Abstract：Cros-domainrecommendation technologyeffectivelyenhances therecommendation performanceof the targetdomain bydeeplyminingandutilizing useful informationfrom otherdomains，providing anefectivesolution tousercold-start problem.However，existingcross-domainrecommendationmethods havelimitations，failing tofinelyexpandimplicitrelationshipsandneglectingpotentialredundantinformationinembeddingvectors，whichrestrictstheperformanceofcross-domain recommendationsystems.Toaddressthis，this paper introduced theintra-domainandinter-domainmeta-pathsaggregation basedcro-domainrecommendationmethod（IMCDR）.Specifically，IMCDR firstlycalculated thefine-grainedsemanticembedding of entities across multiple fields，thereby efectively extending user-userand item-itemrelations.Then，IMCDR generated privatefeaturesandshared featuresforeach nodebasedonintra-domainmeta-pathsand inter-domain meta-paths respectively，andefectivelyintegratedtemtoobainbeterqualityembeddngvectors.Experimentalresultsacrosstecossdomain recommendation tasksdemonstrate that IMCDR significantlyoutperforms in terms of precision andoverallperformance.

Keywords：cross-domainrecommendation；usercold-start problem；graph neural network；heterogeneous informationnetwork；meta-path

0 引言

在快速变革的数字化时代，各种应用不断涌现，海量产品、新闻和视频等内容不断发布和更新，网络信息和数据呈爆炸式增长，互联网用户难以在信息海洋中找到真正有价值和感兴趣的内容[1，2]。（剩余24084字）