CF框架下物質(zhì)擴(kuò)散算法改進(jìn)策略研究

2014年11月30RecommenderCollaborativeContent-basedNetwork-basedHybridSimilarity-basedDiffusion-basedDimensionalityreductionSocialMeta劉建國(guó),周濤,汪秉宏個(gè)性化推薦系統(tǒng)研究進(jìn)展[J].自然科學(xué)進(jìn)展2009,19(1):1-LinyuanLü,Matú?Medo,etal,Recommendersystems,PhysicsReports,Volume519,Issue1,October2012,Pages1-49,ISSN0370-1573.Diffusionvs.Collaborative isakind 1l a ll

ai

assalton

aa

1kk2

l

lKeyfactorsofitemSimilarityofitemαandβdepends?kαand?Numberofcommonusersofαand?InterestdiversityofcommonCFusuallycapturestheformertwoMDcapturesallthethreeDiffusion-likeDiffusion-likeMD-MassDiffusion:

1

l HC-HeatConduction:

1

a

l

l

l

Biased

l

aakPreferential

l r r

akPartInformationfilteringviabalanceddiffusiononbipartitenetworksBalancedDiffusion

ak l

kk l [1]D.-C.Nie,Y.-H.An,Q.Dong,Y.Fu,T.Zhou,Informationfilteringviabalanceddiffusiononbipartitenetworks,PhysicaA(2014),/10.1016/j.physa.2014.11.018OptimalparameterComparisonofDegreeofitemsinrecommendingPartInformationFilteringviastrengtheningthecorrelationofsmalldegreeobjects a

a

l

l DistributionofCorrelationbetween

c(Correlation

c(CorrelationThecorrelationobeypower-lawdistributionwhichmeansmostofthecorrelationvaluebetweenobjectsaresmall.

l

againstobject a

a

ll

l σ=σ=Table2.PerformancesonMovieLensandTable3.PerformancesonMovieLensondifferentTable4.PerformancesonNetflixondifferentPartAttemptonnonlinearhybridofCFandAttemptal

a

l

l

l Bettermodel-

alal

l Table6.ComparisonofdifferentDataPartFour:Directedsimilarityinkk k a w RankingDiffusion-likealgorithmsareessentiallyWecangetbetterrecommendationresults,?Increasingthesimilarityofsmall-degree?ModifyingthesimilarityofCFinspiredbydiffusion-likesimil