文本特征選擇算法的研究.pdf

1、遼寧師范大學(xué)碩士學(xué)位論文文本特征選擇算法的研究姓名：楊榮杰申請(qǐng)學(xué)位級(jí)別：碩士專業(yè)：計(jì)算機(jī)應(yīng)用技術(shù)指導(dǎo)教師：任永功201204一文本特征選擇算法的研究—————————————————————————————————————————————————一。——ResearchonFeatureSelectionAlgorithmforTextAbstractTextclassificationTechnologyclassifiedthema

2、ssiveunclassifiedintormationontheIntemetviapreviouslydefinedcategory，toimprovetheutilizationrateofinformationandveri妙thevalueofinformationforresearchersTextclassificationisaprocesswhichcont1jnstextCOrpus，textpretreatment

3、，textpresentation，charactel’dimensionalityreduction，struc：ureclassifierandtestAmongthem，thecharacterdimensionalityreductioniSthemostimpor：antoneoftheSIXstepsInimprovingtheaccul’acyofclassification，savingspace，reducetheca

4、lculationtime—complexity，dimensionalityreductionplayaveryimportantrole1herearetwomajorwaystoreducecharacterdimensionally：featureextractionandtcatureselectionFeatureextractionabstractsanewfeatm‘esubsetbytherelationshipoft

5、hemappingfilnctionfromtheoriginalfeaturesThecharactersbetweenfcaturesarem3reindependentinthisspecificcharacterdimensionandit’SmucheasiertoidentifvvarioustextsFeatureselection，amongthetotalfeaturescluster，selectsthemorede

6、signablefeaturesas：heuniquesubsetofclassifiedtrainingbyusingtheFeatureweightscalculationformulawithvariousfeaturesDependondifierentperspectives，featureselectioncouldbedefineddifferently，amongthem，therearetwoonesnamed‘‘un

7、dersupervision，’and“nosupelmsion”whichcouldbeidentifiedbywhetheritgettheclassmarkerThispapmincludestxvoalgorithms：Firstofall，mRMRReliefFfeatureselectionalgorithmwhichisbased011ReliefFalgorithmThealgorithmtakesprobability

8、insteadtomakeupthelackofdifferencesoft11efeaturemeasurement，andcomeupwithaHewdifierencefunctionThefeaturesextractedfromthefunctionmakesthecorrelationwithinclassandthedifferencebetweenclassesmoFesignificantThealgorithmals

9、oconsidersthewordcorrelationWordcorrelationisnconsiderationofnotonlychoosingthewordwithcloserelationtothecharacteristiconesbutalsoeliminationofcharacteristicredundantThroughthecomparisonexperimentsofthethr(ealgorithms，th

10、eresultverirythatthetestalgorithmprovidesamoreeffectivefeaturetotextclassificationinthispaperSecondlyimprovementforconventionalIGalgorithm：atextfeatureselectionmethoj(TDpIG)basedoninformationgainSelectingdataclusterbyfea

11、turecategoryteduceth：influenceofunbalanceclusterforthefeatureselectionSecondstep，byusinlzfeatures‘a(chǎn)ppearingprobabilitycalculategainvalueofinformationtolowerthedisturbanceoflOWfrequencywordsforfeatureselectionFinally，anal