數(shù)據(jù)挖掘畢業(yè)論文外文翻譯

1、外文文獻原文外文文獻原文DataMining2UsagescenariosDataminingiswidelyusedinarangeofscientificdisciplinesbusinessscenarios.SomenotewthyexamplesincludefindingsintheareasofdatabasemanagementmachinelearningBayesianinferenceknowledgegainfe

2、xpertsystemsfuzzylogicneuralwksgeicalgithms.Examplesineverydaybusinessscenariosincludedatabasemarketingfairlinespaneldataresearchaswellasthecreationofcustomizedtradepublicationsbasedonsubscriberdatafhundredsofdifferentus

3、ergroups.FrawleyPiatetskyShapiro(Frawleyetal.99)offeradetailedoverviewoffurtherareasofusage.Grossmarginanalysisisanotherinterestingfieldofresearchindatamining.Withthehelpofmoderncostaccountingsoftwarecompaniescanperfmmul

4、tidimensionalanalysisonindividualincomeitems.Fig.2listsafewsamplequestionsrelatedtothistopic.Duetothenumerousreferenceobjects(e.g.productscustomerssaleschannelsregions)theresultingnumberofobjectsthatneedtobeexaminedcontr

5、ollersrequiremethodsthatautomaticallyidentifydatapatterns.Inthiscasethesepatternsareacombinationofattributevalues(e.g.“DIYstes”“powerdrills”inFig.1)aswellasmeasures(e.g.grossmargin).Acompanythatdevelopsadataminingprogram

6、mustalsoconsiderthelargevolumesofdatainvolved.Eveninasizecompanyfexampleitiscommonthatseveralhundredthousitemsflowintoamonthlyincomestatement.CaseBasedReasoning(CBR)isoneinterestingexampleofhowdataminingmachinelearningco

7、uldwktogether.CBRcomponentsattempttotracecurrentquestionstoproblemsthathavealreadybeensolvedinthepast.Helpdeskswhichassistinclarifyingthequestionsacustomerhasaboutpurchasedproductsareonepracticalusageofthistypeofprocedur

8、e.Whilesomecompaniesusehelpdeskstosuppttheirtelephonehotlinesothersgivetheircustomersdirectaccessthrougharemotedatatransfer.Dataminingcanbeveryvaluableinthiscontextbecauseitconsolidatestheinfmationgatheredinthoussofindiv

9、idualhisticalcasesintokeyfindings.Theadvantageofthisprocedureistheshterprocessofsearchingfprecedentswhichcanbeusedtoanswerthecurrentcustomer’squestion.3MethodsTherearemanydifferenttypesofmethodstoanalyzeclassifydata.Some

10、commonmethodsincludeclusteranalysisBayesianinferenceaswellasinductivelearning.Clusteranalysissignificanttakeacompletelydifferentapproach.Initiallyeachobjectislocatedinitsowncluster.Theobjectshoweverarethencombinedsuccess

11、ivelysothatonlythesmallestlevelofhomogeneityislostineachstep.Wecaneasilypresenttheresultinghierarchyofnestedclustersinasocalleddendrogram.3.1.2ConceptualclusteringAsdescribedabovetraditionalfmsofclusteranalysiscanidentif

12、ygroupsofsimilarobjectsbutcannotdescribetheseclassesbeyondasimplelistoftheindividualobjects.Theobjectiveofmanyusagescenarioshoweveristoacterizetheexistingstructuresthatareburiedamongthevolumesofdata.Insteadofrepresenting

13、objectclassesthroughsimplylistingtheirobjectsconceptualclustersintentionallydescribethemusingtermswhichclassifytheindividualobjectsthroughrules.Agroupoftheserulesfmsasocalledconcept.Abasicexampleofaconceptisaprogramthata

14、utomaticallylogicallylinksindividualattributevalues.Advancedsystemscanevenestablishconceptsconcepthierarchieswithclassificationrules.Thedifferentconceptsinpartitionalmethodsofconceptualclusteringcompetewitheachother.Ulti

15、matelywehavetochoosetheclusteringconceptthatbestmeetstheperfmancecriteriafaspecificmethod.Someperfmancecriteriaincludethesimplicityoftheconcept(basedonthenumberofattributesinvolved)thediscriminatypower(asthenumberofvaria

16、blesthathavevaluesdonotoverlapbeyondthedifferentobjectclasses.)Similartotraditionalclusteranalysistherearealsohierarchicaltechniquesthatfmclassificationtreesinatopdownapproach.Asdescribedabovethebestclassificationinterms

17、ofperfmancecriteriawilltakeplaceoneachlevelofthetree.Theprocessendswhennofurtherimprovementispossiblefromonetree4CriticalfactsThefollowingsectionoutlinessomeproblemsassociatedwithdatamining.Inouropinionthesecriticalfacts

18、fsuccesswillfmthefoundationffutureresearchdevelopment.4.1EfficiencyofalgithmsRegardingtheefficiencyofdataminingalgithmsweshouldconsiderthefollowingaspects.Calculationtimesareakeyfact.Ifthecalculationtimesofalgithmsgrowfa

19、sterthanthelineardependencyofthesquarednumberofdatarecdstobesearchedwecouldassumethattheywouldnotbesuitableflargerapplications.Wecanimprovecalculationtimesbylimitingthesearchareathroughuserinputreducingthesearcheddatavol

20、umethroughtargeted(e.g.userbased)ioncompression.Recentdevelopmentsshowthatthecalculationtimeofalgithmswillbecomelessrelevantduetotechnicaldevelopments(e.g.fasterprocesssparallelcomputers).Thealgithmsmustberobustenoughtod