改进的单类协同过滤推荐方法.pdfVIP

ISSN 1673—9418 CODEN JKYTA8 JournalofFrontiersofComputerScienceandTechnology E—mail：fcst@vip．163．corn 1673-9·4182／014／08(10)-·1231-08 http：／w／ww．ceaj．org Tel：+86．10 doi：10．3778／j．issn．1673—9418．1403057 改进的单类协同过滤推荐方法 王 鹏，景丽萍 北京交通大学 交通数据分析与挖掘北京市重点实验室，北京 100044 ImprovedOne—ClassCollaborativeFilteringforRecommendationSystem WANG Peng，JINGLiping BeijingKeyLabofTrafficDataAnalysisandMining，BeijingJiaotongUniversity,Beijing100044，China +Correspondingauthor：E-mail：lpjing@bjtu．edu．cn WANG Peng，JING Liping．Improvedone-classcollaborativefliteringforrecommendation system．Journal ofFrontiersofComputerScienceandTechnology,2014，8(10)：1231—1238． Abstract：Thesparsityofdataandthelackofnegativesampleshaveabadinfluenceontheresultofmatrixfactor— izationbasedone-classcollaborativefilteringapproach，suchasfeaturesextractedarenotobviousandlow differen- tiationbetweentrainingresults．Thispaperproposesanalgorithm tochoosepositiveandnegativesamplesbyconsi— deringbothitem similarityanduseractivity．W iththisalgoritmh ，morepositivesamplescanbeaddedviatheitem similarityandmorenegativesamplescanbeaddedwiththeaidoftheuseractiviyt．Inthiscase，whenmartixfactor— izationisusedinone—classcollaborativefilering，theinfluenceofthesparsiyt andthelackofnegativesamplesCna beeffectivelyreduced．A seriesofexperimentsshow thatthenew algorithm Cna addpositiveandnegativesamples moreaccurately,reducethesparsity,andimprovetherecommendationaccuracy． Keywords：matrixfactorization；one—classcollaborativefiltering；sparsiyt；positiveandnegativesamples 摘 要：在使用矩阵分解方法解决单类协同过滤问题时，数据的稀疏性以及 负样本的缺乏会导致分解特征提 取不明确，训练结果区分度低等诸多弊端。针对此问题提出了一种综合考虑物品相似度以及用户活跃度的正 负样本选择算法，根据物品相似度向原始数据中添加一定正样本，同时根据用户活跃度向每个用户添加不同 数量的负样本，从而减小了稀疏性和缺少负样本对使用矩阵分解方法解决单类协同过滤问题的影响。实验结 TheFundamentalResearchFundsfortheCentralUniversitiesofChinaunderGrant