融合信息增强和兴趣演化的个性化推荐

打开文本图片集

中图分类号：TP391 文献标识码：A 文章编号：2096-4706（2025）15-0122-07

Personalized Recommendation Integrating Information Enhancement and InterestEvolution

LIANG Songyang， YANG Zexin （GuangdongUniversityofPetrochemical Technology，Maoming 525ooo，China）

Abstract： The sequence prediction models canefectivelycapture dynamic changes inuser behaviors，and can provide more personalized predictions forrecommendation systems.However，existingmodels focus onmodelconstructionandoverlook the isueof dataimbalance.DpInterestEvolutionNetwork with InformationEnhancement IE-DEN）isdesigned forthis purpose， andthe modelintegrates informationenhancementmethodsbasedontheDENarchitecture.Specificaly，theoperationistocreate anembedding vectorforeachuserandcalculatethe Hadamard product withthesequence samples toupdate the spatial distance between thesamples.The enhanced samples are inputito structures such as interest extraction layers，whichcan beregarded as hyperplaeclassfcationmodels topredictuserinterestpreferences.Theexperimentsshowthat IE-DIENimprovesAUCandMCC indicatorsby more than 5% onboth datasets.Especially， it performs better on datasetswith imbalanced data.

Keywords： sequence prediction; information enhancement; interest evolution; data imbalance

0 引言

基于序列预测的推荐模型（RecommendationModel Based on Sequence Prediction，RM-SP）是一种利用用户历史行为序列数据，预测用户未来可能的兴趣或行为倾向的算法模型。（剩余9850字）