基于模式相似的子空間聚類算法研究.pdf

1、學(xué)號(hào)；』盟韭婭Q!三一廿婁目————衛(wèi)蛆L————一密扭——』巳—一Ⅲ一———giL————一河I辜大孳碩士學(xué)位論文基于模式相似的子空間聚類算法研究f埕指導(dǎo)教師姓名—雎墨蓮J岫刳豇坷漣盤生讓簋扭厘擅息蘭五蘭EL—直jL直上L鏖』LL芏—三垃墅L一——申請(qǐng)學(xué)位級(jí)別_王壁殖——專業(yè)g稱—丑擅皇值顯亟自L論文提交日期t』地吐絲且論文粹辯日期，』盟臼曲且Ⅱ且學(xué)位授予單位和日新』L盤』二芏————型籩j三生_且一瞢辯委員畬主常：—圭且王一論文評(píng)閱

2、人三臼【5L、jUl2L2005年3月中國(guó)南京河海人學(xué)碩士論文AbstraCAbstractClusteringanalysis，whichisanimportantdataminingproblem，groupsthedataintoclassesorclustemsothatobjectswithinaclusterhavehighsimilarityincomparisontooneanotherbutareverydissimi

3、lartoobjectsinotherclustersIthasbeenwidelyusedinnumerousapplications，includingpaRemrecognition，dataanalysis，imageprocessing，recommendsystemandelectroniccommerceInthispapergeneralcategorizationsofclusteringmethodsarediscu

4、ssedandthenthemainclusteringmethodsareanalyzedindetailThecomparisonofthemainmethodsismadeTraditionalclusteringmethodsCallworkefftcientlyinlowdimensionaldataInhiighdimensionaldatahowever、efficiencyandeffectoftraditionalcl

5、usteringmethodsarenotwellbecauseofdatasparsitydistancesimilarityandmoreoutlierinthedataTechniquesforclusteringhighdimensionaldataincludebothfeaturetransformmion(dimensionsimplification)andsubspaceclustering(featureselect

6、ion)techniquesWeillustrateandmakecomparisonseveralsubspaceclusteringalgorithmproposedinrecentyearsinchapter2111eclusteringalgorithmbasedonpatternsimilarity—pClusterisoneofsubspaceclusteringDifferfromtheclusteringmodelsba

7、sedondistance，pClustermodeldefinesthaitwoobjectsaresimilariftheyexhibitacoherentpaRemonasubsetofdimensionsThepurposeofpClusteralgorithmistevealthiskindofpaRemsimilarityamongobjectsWediscussandimplementthealgorithmandprop

8、oseallimprovementalgorithmforitsshortcominginchapter3，MCAS(MaximumCoherentAttributeSem)pruningbyobjectblockisusedtoproneinvalidMCASsinsteadofsymmetricMCASpruningAtthesametimeweenumerateattributepaironeverybranchofprefixt

9、reeandcalculateintersectionofMCOS(MaximumCoherentObjectSets)onobjectsonthebranchTheexperimentprovesthattheefficiencyspaceoverheadandeffectoftheimprovementalgorithmarebetterthanoriginalalgorithmAtlastaprototypeofrecommend