一种面向在线交易场景的序列推荐方法

打开文本图片集

中图分类号：TP391.3；TP181 文献标识码：A 文章编号：2096-4706（2025）15-0087-08

Abstract：The sequence recommendation system based on online transaction scenarios aims to predict the next item that users maybe interested inbasedontheusers existingaccesshistory，andhasbecomean indispensable partofmajoraplication malls.However，themainprobemfacedbycurentresearchinthisfeldisthatthesaleoftheorigialdatasetisisuficent， which makesthemodelunable tobeeectivelylearedanditisdificult toaccuratelymineuserintentions.Tothisend，a sequence recommendationmethodbasedonMultipleEnhanced Contrast （MEC）frameworkis proposed.This method integrates the idea ofContrastive Learning.Firstly，theoriginal data is enhancedbyvarious typesof datato generatealarge amountof simulationdata.Secondly，theself-supervisedsignalofsmilarityrepresentationbetweensimulationdataisstablished.Finaly theMECmodeliscombinedwiththepairwiseranking lossforlearning.Extensiveexperimentsontwopublicdatasetsshowthat theproposedmethodhassignificantlyimprovedsequencerecommendationperformance inonlinetransactionscenarioscompared with the current baseline method.

Keywords：sequence recommendation; data augmentation; MEC framework; self-supervision; Contrastive Learing

0 引言

随着移动应用和服务人群数量的激增，为了避免信息过载，更好地服务于各类用户，序列推荐系统作为机器学习领域的新宠引发了研究人员的广泛关注[1-3]。（剩余13532字）