推荐系统中的隐私保护：技术演进、评估框架与隐私效用协同发展

打开文本图片集

关键词：隐私保护；推荐系统；差分隐私；加密技术；联邦学习；k-匿名中图分类号：TP309.2 文献标志码：A 文章编号：1001-3695（2025）12-001-3521-13doi：10.19734/j.issn.1001-3695.2025.04.0135

Evolution of privacy protection in recommendation systems ： technical advancements，evaluation frameworks，and synergistic development of privacy and utility

Wang Bin 1a，1b，2 ，Wang Changla，b，Lyu Linghuila1b，Liu Yutao 1a，1b ， Zhang Lei 1a，1b† （204号 （1.aHerceKbofele&ossnsiKbi gationTecholog&EqientEgininThogodoffotio&ElectocTcoogysiUniesitysiHei longjiang15407，China；2.Science&TechnologyDept，Jiamusi University，JiamusiHeilongjiang，China）

Abstract：TherapiddevelopmentoftheInternetandmobiletechnologies hassignificantlyimproveddatacolectionandprocessing capabilities，enabling recommender systems todeeply mineuser behavioranddeliver more precise personalized services，improvinguserexperience.However，theaccumulationofvastamountsofdataand thewidespreaduseofrecommender systemshavealsointroducedsevereprivacyleakagerisks.Achievingeficientrecommendationswhileprotectingprivacyhas become a key research focus.Thispaperreviewed the evolution of privacy protection methods inrecommender systems and identifiedkeyriskpoints intherecommendationprocess.Itanalyzedthedesignprinciplesandapplicationcharacteristicsof exi-sting methodsandcomparedthediferentapproachesusing evaluationmetrics forprivacyprotectionstrengthandrecommendationefectivenessItsummarizedtheadvantagesandlimitationsof thesemethods，proposedoptimizationdirections，and discussedposiblefutureresearchpaths.Theseresultsprovidetheoreticalandpractical guidancefortheaplicationof privacy protection mechanisms in recommender systems.

Key words：privacyprotection；recommendersystem；diferential privacy；encryptiontechnology；federated learning；kanonymity

0 引言

随着信息技术的迅猛发展，推荐系统通过从海量信息中筛选出符合用户兴趣需求的内容来提升用户体验[1]，广泛应用于电子商务[2]、社交网络[3]、新闻媒体[4]和互联网医疗[5]等领域。（剩余45447字）