基于超图和分层频谱滤波器的序列推荐模型

打开文本图片集

中图分类号：TP391.3 文献标志码：A 文章编号：1001-3695（2025）07-013-2018-07

doi： 10.19734/j. issn.1001-3695.2024. 12.0488

Abstract：Toaddress thedata sparsity problem prevalent insequence recommendationandthe noise isue caused byunexpected interactions between users and items，this paper proposed asequence recommendation model basedon hypergraph and hierarchical spectralflters（HYFTRec）：HYFTRec introducedthe hypergraph structure intosequence recommendation， captured thecomplex higher-orderrelationshipsbetween usersanditems through thehypergraphembedding module，and atthe sametimeutilizedthehierarchicalspectralfilterforeffctivedenoisinginthefrequencydomainforefectivedenoising，which improved theaccuracyandrobustnessofrecommendation.Inaddition，themodel incorporatedacomparative learning framework tooptimize the characterizationabilityof userbehaviorsequences.Through multipleexperimentalvalidationsonthree public datasets，HYFTRecdemonstratesadvantages intermsof keymetricshitrate（HR）and normalizeddiscountcumulative gain（NDCG），which significantlyoutperformsexisting sequencerecommendation models.Compared with the benchmark modelFMLP，HYFTRec improves 10.7% ， 10.8% ，and 7.6% in HR@ 10 metrics and 12.5% ， 13.7% ，and 7.3% in NDCG@10 metrics，respectively.These results verify the validity and superiority of the proposed model.

Key words：hypergraph；hierarchical spectral filter；contrast learning；sequence recommendation

0 引言

近年来，随着推荐系统的广泛应用，序列推荐逐渐成为研究的热点领域[1＼～4]。（剩余17639字）