ClusterAnalysisTableofContents�AboutClusterAnalysis�AreaofApplication�AdvantagesComparedwithTraditionalMethods�ApplicationFlow�CaseStudyIntroductionofClusterAnalysisIntroductionofClusterAnalysis♦Ageneralquestionfacingresearchersinmanyareasofinquiryishowtoorganizeobserveddataintomeaningfulstructures,thatis,todeveloptaxonomies.♦Thetermclusteranalysis(firstusedbyTryon,1939)actuallyencompassesanumberofdifferentclassificationalgorithms.♦Clusteranalysismethodsaremostlyusedwhenwedonothaveanyapriorihypotheses,butarestillintheexploratoryphaseofourresearch.Inasense,clusteranalysisfindsthe"mostsignificantsolutionpossible.Sosignificancetestingisnotappropriate.DefinitionDefinition♦♦ClusterAnalysisisamultivariateanalysistechniqueClusterAnalysisisamultivariateanalysistechniquethatseekstoorganizeinformationaboutvariablessothatseekstoorganizeinformationaboutvariablessothatrelativelyhomogeneousgroups,or"clusters,"thatrelativelyhomogeneousgroups,or"clusters,"canbeformed.canbeformed.♦♦TheclustersformedwiththisfamilyofmethodsTheclustersformedwiththisfamilyofmethodsshouldbehighlyinternallyhomogenous(membersshouldbehighlyinternallyhomogenous(membersaresimilartooneanother)andhighlyexternallyaresimilartooneanother)andhighlyexternallyheterogenousheterogenous(membersare(membersarenotnotlikemembersoflikemembersofotherclusters.otherclusters.分为按分为按casecase聚类和按变量聚类，按聚类和按变量聚类，按casecase聚类应用更普遍。聚类应用更普遍。OutputofaclusteranalysisDendrogram---byvariableThetwodimensionsoftaskandpeopleskillsalsoemergefromthisanalysis.ThedifferencefromFactorAnalysisisthatyoucanseewhichvariablesareclosertotheothers,basedonwhichlinkfirst.AreaofApplicationI(inScience)♦wheneveroneneedstoclassifya"mountain"ofinformationintomanageablemeaningfulpiles,clusteranalysisisofgreatutility.�Medicine:clusteringdiseases,curesfordiseases,orsymptomsofdiseases�Psychiatry:thecorrectdiagnosisofclustersofsymptomssuchasparanoia,schizophrenia,etc.isessentialforsuccessfultherapy�Archeology:researchershaveattemptedtoestablishtaxonomiesofstonetools,funeralobjects,etc.byapplyingclusteranalytictechniquesTheproximitiesaredistancesfor20species.聚类分析的应用II（市场研究）�市场细分�寻找目标消费群体；�划分产品的细分市场；�描述各细分市场的人群特征�购买者行为研究�试销市场的选择�竞争策略分析�......图图1.1.产品满意度测试产品满意度测试((a)a)同质型同质型((b)b)分散型分散型((c)c)群组型群组型价格价格价格功能功能功能目的：考察不同组群的人口、经济特征图图2.2.试销市场的选择试销市场的选择目的：研究消费者特征，确定出目标消费群聚类分析同传统研究方法的对比优点缺点�不易得到目标消费群体综合特征�测试方法简单，在测试消费者对产品的需求时可粗略地分析出目标消费群体特征传统测试方法聚类分析方法�可以方便地得到目标消费群体的综合特性，包括其生活方式、收入背景、教育背景等Case:WhiskyClassified20001.It'snotaRegionalClassification!TheconventionalwaytoclassifyScotchmaltwhiskiesisbyregion-Highland,Lowland,Speyside,IslayandCampbeltown.Butknowingwheretheyaremadedoesn'texplainhowtheytaste.NotallIslaymaltstastelikeaclassic,smokyslay!SomeSpeysidersarelightanddelicate,whereasothersarerichandfruity.2.It'snota"Quality"Classification!�Somewhiskywriterstrytoassessthe"quality"ofsinglemaltwhiskies-theyawardmarks-out-of-tenandconstructleaguetablesof"top"whiskies.Theyareoftenlookingfordepth,balance,layeredcomplexityandlengthoffinish-crite...
