推荐系统中若干关键问题研究

《推荐系统中若干关键问题研究》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、南京航空航天大学博士学位论文推荐系统中若干关键问题研究姓名：李涛申请学位级别：博士专业：计算机应用技术指导教师：王建东20081001推荐系统中若干关键问题研究（4）信息安全和隐私保护是数据挖掘领域的热点之一。推荐系统需要收集用户兴趣喜好等相关数据，在一定程度上涉及了用户的个人隐私，因而推荐系统中的隐私保护也开始受到研究人员的关注。本文提出了一种基于随机扰动的隐私保护推荐算法。算法在用户数据收集过程中采用随机扰动技术，并使用非负矩阵分解对数据进行处理，从而形成隐私保护功能，并在此基础上产生推荐。通过分析及实验表明，算法在保护

2、用户个人隐私的基础上，能够产生具有一定精确度的推荐结果，以满足推荐系统的需要。关键词：推荐系统，协同过滤，数据挖掘，聚类，隐私保护II南京航空航天大学博士学位论文AbstractWiththedevelopmentoftheInternet,thewebprovidesmoreandmoreinformationforusers,whilethestructureofthewebhasalsobecomemoreandmorecomplex.Thissituationhasmadeitsubstantiallymoredif

3、ficultforuserstofindtheinformationtheyneedfromthevastamountofmaterialsavailableontheInternet.Therecommendersystemprovidesinformationfilteringforauserbypredictingtheparticularuser'spreference,anditcanapplyknowledgediscoverytechniquestomakepersonalizedrecommendations

4、tohelptheuserquicklyfindthedesiredinformation.Atthesametime,therecommendersystemcanenableenterprisestoachievetheobjectiveofpersonalizedmarketing,whichcanimprovesalesandgeneratemoreprofits.Inaddition,withthepopularizationofpersonalizedservice,therecommendersystemisw

5、idelyusedonagrowingnumberofwebsites,especiallyintheE-Commerceplatform.Becauseofitsgreatpotentialfordevelopmentandapplications,therecommendersystemhasbecomeanimportantresearchareainwebintelligenttechnologiesandattractedsignificantattentionfromresearchers.Althoughthe

6、developmentofrecommendersystemhasbeensuccessfulinbothresearchandapplications,anumberofchallengingresearchproblemsstillexist.Toaddressthesechallenges,thisdissertationexploresandstudiessomekeytechnologiesoftherecommendersystem,suchasthedesignofnovelalgorithmswithbett

7、errecommendationqualityandenhancedprivacyprotectiontechnology.Inparticular,dataminingandmachinelearningtechniquesareincorporatedintotherecommendersystem.Technologiesforenhancedreal-timerecommendation,improvedrecommendationqualityandstrengthenedprivacyprotectioninth

8、erecommendersystemareinvestigated.Themainresearchresultsofthisdissertationareasfollows:First,theperformanceofcollaborativefilteringsystemsdegrade