基于類別均衡的文本分類算法研究.pdf

1、代號一】o捌分類號1只靼!學號～∞22蝗j靼密級登五而簧它手辭校六警碩士學位論文題c、t，目基于類別均衡的文本分類算；丟研究l瀚魍r曲旦J9熙騏￡她￡go啦曼艇蚋舷Qn俑熙量曼￡4Q魏￡熟宴gQ強H9mQg!縣jz系騏Q駔作者姓名鄭偉指導教師姓名、職務姜建國教援學科門類王黌學科、專業(yè)謎算杰凡虞用技術論文提交日期皇OQ年1月ABSTRACTAutomatictextclassificationisdefinedasthetasktoass

2、ignpredefinedcategorylabelstodocumentsVectorspacemodel(VSM)isthemodelthatisusedwidelyinthe1a唱escaletextdisposalToimprovemeclaSs湎cationperfomlance，tJlispaperdoesresearchandmlpmVesontheclassicalalgorithmofcalculatingthete響

3、weightinVSMFunhermore，acategoryhomogenizingbasedknearestneighbors(kNN)algorithmfortextcategorizationisproposedTheimportanceofterlnsintextcategorizationandthecapabilityofrenectingtextcontemoftermslieindiffbrentpositionint

4、extareconsideredintheimprovementonalgorithmofcaIcuIatingthete咖wei曲tInformationentropyandpositionoftermsareintroducedinthisapproachbasedonTFIDFThisalgorithmremedyTFIDF’sdef毫ctsthatdistributionproponionandinformationofposi

5、tionoftermisneglectedThetminingsetisorganizednewlywithcategoryasunitinthekNNalgoritllmbasedoncategoryhomogenizationMeanwhile，similarsmallcategorieslessthanthet11resholdarecombinedint0newbigcategoriesa11dassub—categorieso

6、fnewbigcategories，bigcategoriesmorethanthethresholdareseparatedintOsOmesmaHcategOriesandthemassub—categories0fbigcategOriesThecenterOfsub—categoriesisconsjderedasdotofcatego“esThedocumentjscategorizedbasedonsumofcosinedi

7、stallcebetweenitanddotofcategoriesDistributionofcategoriesishOmogenizedintheorganizedtrainingsetTheproblemsofMulti—peakdistribution，overlapboundaryandsmallcatego“esareneglectedareresolvedbycategorizingincategoryhomogeniz