电子商务推荐系统中协同过滤瓶颈问题研究

《电子商务推荐系统中协同过滤瓶颈问题研究》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、合肥工业大学博士学位论文电子商务推荐系统中协同过滤瓶颈问题研究姓名：李聪申请学位级别：博士专业：管理科学与工程指导教师：梁昌勇20090401摘要合肥工业大学博士学位论文从而消除了传统方法在每次进行推荐计算时无法避免的扫描全体项目空间的计算耗费，有效改善了可扩展性；同时，由于这种增量更新机制保证了在推荐运算中能够使用到最新的用户评分数据，因此使得推荐服务可以适应用户兴趣偏好的动态变化，从而弥补了传统的离线计算项目相似性方法难以反映用户兴趣漂移的不足。(5)在本文提出的上述理论和方法基础上，设计并实现了一个电子商务协同过滤原型系统ECRec(E．CommerceReco

2、mmendersystem)，该系统具有良好的可移植性、可维护性及开放式架构(openarchitecture)特征。关键词：电子商务推荐系统；协同过滤；稀疏性；冷启动：可扩展性合肥工业大学博士学位论文ABSTRACTWiththerapiddevelopmentofIntemetandE·commerce，humansocietyhasbeenstepintoinformationera．ThedevelopmentpotentialofChineseE—commerceisenormous，anditkeepsacontinuouslyhigh—speedincr

3、easing．PeoplecallenjoythehappinessandconvenienceofpurchasingproductsviaE．commercewebsitesathome．However’thetremendousproductscategory,whichsuppliedbyE—commercewebsites，brings“informationoverload”tousers．Hence，E。commercewebsitesfacesaseriousproblem：howtorecommendappropriateproductsforbro

4、wsinguserstoovercomethedetrimentaleffectsofinformationoverloadandpromotemoretransactionsforboostingthesalesofwebsites?E．commercerecommendersystemsareoneschemetosettleinformationoverload，andonetechniquetorealize“one—to—one”strategyofE-commercewebsites．Ithasbeenappliedinmanylarge-scaleweb

5、sitesbybeingtreated越ahelpfulpartofcustomerrelationshipmanagementforthewebsites．Collaborativefilteringisthemostsuccessfulandwidelyusedrecommendationalgorithmi11E．commercerecommendersystemscurrently．However，thereexistsomebottleneckproblemsincollaborativefiltering，such硒sparsity,cold-starta

6、ndscalability．Thesebottleneckproblemslimitthedevelopmentofcollaborativefiltering，henceweshoulddeeplystudyontheproblems．Themainresearchworksofthispaperare签follows：(1)Onthebasisofacomprehensiveoverviewontheresearchofcollaborativefilteringathomeandabroad，asummaryontheboaleneckproblemsofcol

7、laborativefilteringisgiven．(2)Toaddressthedrawbacksofitem—rating·predictioncollaborativefilteringalgorithminalleviatingsparsity,namelythatthesearchingofnearestneighborisnotaccurateenoughandthereexistsomeunnecessarycomputingcostinthealgorithm，thenon-targetusersdifferentiatingthe