文献翻译数据类型泛化用于数据挖掘算法.docx

文献翻译数据类型泛化用于数据挖掘算法.docx

《文献翻译数据类型泛化用于数据挖掘算法.docx》由会员分享，可在线阅读，更多相关《文献翻译数据类型泛化用于数据挖掘算法.docx（18页珍藏版）》请在冰豆网上搜索。

文献翻译数据类型泛化用于数据挖掘算法

英文翻译

系别

专业

班级

学生姓名

学号

指导教师

DataTypesGeneralizationforDataMiningAlgorithms

Abstract

Withtheincreasingofdatabaseapplications,mininginterestinginformationfromhugedatabasesbecomesofmostconcernandavarietyofminingalgorithmshavebeenproposedinrecentyears.Asweknow,thedataprocessedindataminingmaybeobtainedfrommanysourcesinwhichdifferentdatatypesmaybeused.However,noalgorithmcanbeappliedtoallapplicationsduetothedifficultyforfittingdatatypesofthealgorithm,sotheselectionofanappropriateminingalgorithmisbasedonnotonlythegoalofapplication,butalsothedatafittability.Therefore,totransformthenon-fittingdatatypeintotargetoneisalsoanimportantworkindatamining,buttheworkisoftentediousorcomplexsincealotofdatatypesexistinrealworld.Mergingthesimilardatatypesofagivenselectedminingalgorithmintoageneralizeddatatypeseemstobeagoodapproachtoreducethetransformationcomplexity.Inthiswork,thedatatypesfittabilityproblemforsixkindsofwidelyuseddataminingtechniquesisdiscussedandadatatypegeneralizationprocessincludingmergingandtransformingphasesisproposed.Inthemergingphase,theoriginaldatatypesofdatasourcestobeminedarefirstmergedintothegeneralizedones.Thetransformingphaseisthenusedtoconvertthegeneralizeddatatypesintothetargetonesfortheselectedminingalgorithm.Usingthedatatypegeneralizationprocess,theusercanselectappropriateminingalgorithmjustforthegoalofapplicationwithoutconsideringthedatatypes.

1.Introduction

Inrecentyears,theamountofvariousdatagrowsrapidlyWidelyavailable,low-costcomputertechnologynowmakesitpossibletobothcollecthistoricaldataandalsoinstituteon-lineanalysisfornewlyarrivingdata.AutomateddatagenerationandgatheringleadstotremendousamountsofdatastoredindatabasesAlthoughwearefilledwithdata,butwelackforknowledge.Dataminingistheautomateddiscoveryofnon-trivial,previouslyunknown,andpotentiallyusefulknowledgeembeddedindatabases.Differentkindsofdataminingmethodsandalgorithmshavebeenproposed,eachofwhichhasitsownadvantagesandsuitableapplicationdomains.However,itisdifficultforuserstochooseanappropriateonebythemselves.tochooseanappropriateonebythemselves.Thisisbecausethedataprovidedcannotbedirectlyusedfordataminingalgorithms.Sincemostdataminingalgorithmscanonlybeappliedtosomespecificdatatypes,thetypesofdatastoredindatabasesrestrictsthechoiceofdataminingmethods.Ifcertainkindsofknowledgeneedtobeobtainedusingsomedataminingalgorithms,datatypestransformationshouldbedonefirstandthisiswhatwecalled“thedatatypesfittabilityproblem”fordatamining.Forthetimebeing,thereisnotoolthatcanhelpuserstodothiskindofdatatypestransformation.Inthispaper,wewillsurveyandanalyzethedatatypesfittabilityproblemfordataminingalgorithms,andthenweproposea“datatypesgeneralizationprocess”tosolvethedatatypesfittabilityproblemfortheattributesinrelationaldatabases.

The“datatypesgeneralizationprocess”includingmergingandtransformingphasesisaproceduretotransformthedatatypesofatttributescontainedinrelations（tables）.Inthemergingphase,theoriginaldatatypesofdatasourcestobeminedarefirstmergedintothegeneralizedones.Thetransformingphaseisthenusedtoconvertthegeneralizeddatatypesintothetargetonesfortheselectedminingalgorithm.Usingthedatatypegeneralizationprocess,theusercanselectappropriateminingalgorithmjustforthegoalofapplicationwithoutconsideringthedatatypes.

2.Relatedwork

Asmentionedabove,becausemanydataminingalgorithmscanonlybeappliedtothedatatypeswithrestrictedrange,userspossiblyneedtododatatypestransformationbeforetheselectedalgorithmhasbeenexecuted.Inthispaper,weproposeageneralconceptcalled“datatypesgeneralizationprocess“whichprovideaprocedurefordoingthiskindofdatatypestransformation.Datatypesgeneralizationcanbeseenasapre-processingofdatamining.Ofcourse,otherpre-processingsuchasdataselection,datacleaning,dimension（attribute）reduction,missingdatahandlingmayalsoneedtobeperformedbeforerunningtheselecteddataminingalgorithm.Insummary,thewholeprocessofdataminingistheso-calledKDD（knowledgediscoveryindatabases）,asshowninFigure1.

Figure1:

TheKDDprocessandtheroleofdatatypesgeneralization.

Thereisamajordifferencebetweenthedatatypesgeneralizationprocessandotherdataminingpre-processes.Otherpre-processes（likemissingvaluehandling）areallindependentoftheselecteddataminingmethod.Thatis,theycanbedonewithoutknowingwhatdataminingalgorithmwillbeused.Butitisclearthatdatatypesgeneralizationprocessdependsonthedesiredminingmethod.Thetargetofdoingdatatransformationusingdatatypesgeneralizationistomakethespecifieddatasetsuitablefortheminingalgorithm.Therefore,ifwewanttoachievethisgoal,wemustsurveyboththedatatypesindatabasesandtheirrelationswithvariousdataminingmethods.TheflowofsolvingadataminingproblemwithdoingdatatransformationisillustratedinFigure2.

Figure2:

Solvingdataminingproblemswithdatatransformationdatatypestransformation

Someresearchersproposedhowtogeneralizethedatacontainedinattributesusing"attribute-orientedinduction"whichallowsthegeneralizationofdata,offerstwomajoradvantagesfortheminingoflargedatabases.First,itallowstherawdatatobehandledathigherconceptuallevels.Generalizationisperformedwiththeuseof"attributeconcepthierarchies",wheretheleavesofagivenattribute'sconcepthierarchycorre-spondtotheattribute'svaluesinthedata（referredtoasprimitiveleveldata）.Generalizationofthetrainingdataisachievedbyreplacingprimitiveleveldatabyhigherlevelconcepts.

Infact,datageneralizationusingattributeconcepthierarchiesisakindofdatatypetransformationwhichreducesthenumberofdistinctvaluescontainedinattributes.Wefirstprovideatypicaldescriptionofthedatatypesfittabilityproblemandadatatypesgeneralizationprocesstodefineandsolvethedatatypestransformationproblemforattributes.Hence,datageneralizationusingconcepthierarchiesisincludedintheprocessforperformingspecifieddatatypestransformation.

Anotherrelatedworkisthatsomeresearcherssurveyedabouthowtotransformdataintonumericalvalues.Almostalldata-drivenalgorithmsutilizenumericinputs.Fromacomputerprocessingpointofview,handlingcomputationswithnumbersiseasierandmoreefficient.Therefore,iftheinputvaluesarenon-numeric（e.g.,textstrings）,theyshouldbeintelligentlyconvertedtomeaningfulnumericalvaluesinmanycases.Numericalvaluescanbeseenasadatatypeandtransformingdataintonumericalvaluesisakindofdatatypestransformation.Thestrategiesareincludedinthedatatypesgeneralizationprocessforperformingdatatypestransformation.

3.Analysisofthedatatypesfittabilityproblem

Inrecentyears,duetotheexplosionofinformationandtherapidgrowthofdatabaseapplications,dataminingtechniquesbecomemoreandmoreimportant.Forthisreason,differentkindsofdataminingmethodsoralgorithmshavebeenproposed.However,itisdifficultforuserstochooseasuitableonebythemselveswithoutpriorknowledgeaboutdatamining.Actually,thekindofdataminingmethodsshouldbeapplieddependsonboththecharacteristicofthedatatobeminedandthekindofknowledgetobefoundthroughthedataminingprocess.Hence,thetypesofdatastoredindatabasesplayanimportantroleduringthedataminingprocessandrestrictthedataminingmethodscanbechosenbyusers.Itistruethatallkindsofdataminingmethodscanonlybeappliedtoparticulardatabasessuitableforeachkindandthisiswhatwecalled"thedatatypesfittabilityproblem"fordatamining.Tosolvethisproblem,weneedtoinvestigatetherelationshipsbetweenthecharacteristicsofthedatatobeminedandvariouskindsofdataminingtechniques.Withtherelation-ships,wecanclearlyanalyzethedatatypesfittabilityproblemandfurtherknowwhetherthedatatypestransformationcanbeperformedornot.Hence,analyzingthiskindofrelationshipsisapreparationworkforourdatatypesgeneralizationprocess,whichexplainswhythedatatypesgeneralizationprocesscansolvethedatafittabilityproblem.Wenowillustratetheanalysisasfollows.

3.1Fourkindsofdataformsfordatamining

Dataminingtechniquesususallycanbeappliedtofourkindsofdataforms:

texual,temporal,transactionalandrelationalforms.Differentkindsofdataformsareusedtostoredifferentkindsofdatatypes.Wedescribeeachkindofdataformsinthefollowing:

（1）Textualdataforms:

Textualdataformsareusedtorepresenttextsordocuments.Basically,thiskindofdataformscanbeseenasasetofcharacterswithhugeamount.

（2）Temporaldataforms:

Time-seriesdataisstoredintemporaldataforms.Datathatvarieswithtime（suchashistoricaldata）canbestoredintheformofnumericaltime-series.

（3）Transactionaldataforms:

Forexample,thepasttransactionsofamarketcanbestoredintransactionaldataforms.Eachtransactionrecordsalistofitemsboughtinthattransaction.

（4）Relationaldataforms:

Relationaldataformsarethemostwidelyuseddataformsandcanstorediffierentkindsofdata.Thebasicunitsofrelationaldataformsarerelations（