JournalofChineseComputerSystems201155Vol32No.52011:20100818:(2008AA01A317).:,,1984,,;,,1964,,,IP;,,1979,,.1,2,1,2,21(,210000)2(,100190)Emai:lwangb@fdsp.ac.cn:对视频点播系统中用户行为进行建模和仿真,是研究系统使用状况设计性能优化算法的重要手段.但在以往的研究中,对用户行为建模和仿真都是基于整体历史数据的统计进行的,而在很多情况下,对不同模式的行为采用不同的策略能够更好的提供视频传输服务.本文针对视频点播系统中用户点播行为的特性,以及系统优化策略的需要,提出用户行为时间序列模型和聚类方法,在中国科技大学视频点播系统实际数据基础上进行了仿真测试,结果表明了该方法的可行性.:;;;:TP391:A:10001220(2011)05086704TimeSerialModelandClusteringofUserBehaviorinVideoondemandSystemWANGBingfei1,2,WANGJinlin1,2,LIUXue21(AutomationUniversityofScienceanTechnologyofChina,Hefei210000,China)2(NationalNetworkNewMediaEngineeringResearchCenter,InstituteofAcoustics,ChineseAcademyofSciences,Beijing100190,China)Abstract:ResearchofsimulationoftheuseractioninVODsystemisveryimportantfortheanalyzingofthesystemandthedesigningofalgorithmforoptimizingsystemperformance.Allthepreviousresearchesarebasedonthewholedatasetofuseractions,theycannotreviewpatternoftheseactionseffectively.Anoveltimeserialmodelofuserbehaviorandtheclusteringalgorithmforitispresentedtoresolvethisproblem.ResearchbasedontheUSTCVODsystemshownthatthismethodisfeasible.Keywords:userbehavior;clustering;videoondemandsystem;timeserialmodel1,.,,,,,,,.,.[1]markov,VCR,.[2],Poison,zipf,,.,,,,.,,,,,.,web,,[3]webmarkov,morkov,.,,,,.,,.,,,..2,,.,WebVODP2P,,,.,,1.1Fig.1FrameworkofVODsystem,,,VCR,,,,,,.,,.3,,,,,,,,.,,.,,,.,,,2Fig.2Modelofuserbehavior.,web,,,.2,,.:1.,,.2..3.,.,;,.,:.,,.1.().Action_Type_Set.Action_Type_Set={At1,At2,,Atn}(1)2.().Content_Set.Content_Set={C1,C2,,Cm}(2)3.(Action):TtTypeC.:Action(T,t,At,C)s.t.AtAction_Type_SetCContent_Sett(3)4.(Action_Seq).[0,T]:Action_Seq[0,T]=(Action1,Action2,,Actionn)s.tActioni=Action(Ti,ti,Typei,Ci),0TiTijTiTj(i,jN)(4),,,,,.4,,.,,,,.4.1[0,T]Action_Seq1Action_Seq2,.Action1(t1,0,C1,Type1)Action_Seq1Action2(t2,0,C2,Type2)Action_Seq2,:diff(Action1,Action2)=wtime|t2-t1|T+wt(Type1,Type2)+wc(C1,C2)(5):(x,y)=0(x=y)1(xy)(6),,,.Action_Seq1=(Action11,Action12,,Action1n)8682011Action_Seq2=(Action21,Action22,,Action2m):M(Action_Seq1,Action_Seq2)={(Action1,i,Action2,j)|1im,1jn}(7),,.MaMst.Acontion1iAction_Seq1(Acontion1i,Action2j)MaAcontion2iAction_Seq2(Acontion1j,Action2i)MaMa:diff(Ma)=ni=1diff(mi)n(8)M_Set,:diff(Action_Seq1,Action_Seq2)=Min(diff(Ma))s.t.MaM_Set(9),,M_Set,.4.2,,,,,,,,,,..,:diff(Action_Seq1,Action_Seq2)=1-T0k(C1(t),Type1,C2(t),Type2)dtT(10):k.k:k(C1,Type1,C2,Type2)=wt(Type1,Type2)+wc(C1,C2)(11),,,.,:diff(Action_Seq1,Action_Seq2)=Min(diff(Ma))s.t.MaM_Set(12),:diff=wddiffd+wsdiffs(13)4.3,,,,,,.,,,(DynamicTimeWarpingDistanceDTWD),,,,O(mn),(mn),.,,70%6,90%12,.4.4,.kmean,,,,,,,,.,,,DBSCAN.55.1.2001,(CERNET).,WindowsMediaRealServerG2,.1Table1Statisticalinformationforexperimentdata20080326200805264.9Tbytes85Tbytes156984720001128000ip850028,,,90,,,.2Table2Everytypeusercharacter()()()1[510][1,3]2[1030][1,3]3[1030][3,6]4[3060][1,3],5[3060][3,6],660[1,6],7606,200832620085268695:400,,,,21000,47.2().,(5)(10)(Type1,Type2),.DBSCAN,7,,7,.,,101,,,.1().3Fig.3Distributionofeverytypeofbehavior(:,)5.2,possionpossion,,,.311,2,5,6.,,possion.matlabdfittool.,Nonparametric,4.4Fig.4Fittingofthebehaviordistribution,1,2,4possion,1.possion,,1,.4,,,,1,,,,,,,.20086102008810,,,LFU56Fig.5BytehitratioFig.6Networktrafficratio.56.,,.6,,,,.,,,,,.References:[1]VictorOKL,iFellowIEEE,LaoWanjiun,eta.lPerformancemodelofinteractivevideoondemandsystems[J].IEEEJournalonSelectedAreasinCommunications,1996,14(6):10991109.[2]XianWeiquan,XiangZhe,ZhongYuzhuo.Simulationplatformforuseractionsinthevideoon[J].JournalofSystemSimulation,2001,13(3):221223.[3]HeL.iPredictionmethodofuserbrowsingbehaviorsbasedonclassifiedmarkovchain[J].ComputerEngineering,2008,34(22):3233.[4]VlachosM,HadjieleftheriouM,GunopulosD,eta.lIndexingmultidimensionaltimeserieswithsupportformultipeldistancemeasure[C].Proc.ofthe9thACMSIGKDD2003,Washington:ACMPress,2003,216225.[5]ChesireM,eta.lMeasurementandanalysisofastreamingmediaworkload[C].InUSITS01:Proceedingsofthe3rdConferenceonUSENIXSymposiumonInternetTechnologiesandSystems,USENIXAssociation:Berkeley,CA,USA,2001,11.:[2],,,[J].,2001,13(3):221223.[3].Markov[J].,2008,34(22):3233.8702011