基于神經(jīng)網(wǎng)絡的文本自動分類系統(tǒng)的研究.pdf

1、武漢理工大學碩士學位論文基于神經(jīng)網(wǎng)絡的文本自動分類系統(tǒng)的研究姓名：李淑鵬申請學位級別：碩士專業(yè)：計算機應用技術指導教師：童恒慶20080501AbstractAutomatictextcategorization沁C)hasalreadybecomearesearchfocusinthefieldofinformationprocessing；itreferstothetaskofautomaticallysortingasetofid

2、ocumentsintocategoriesfromapredefinedsetItisacoreoftextminingATCisaneffectivemeansoforganizingandmanagingthemassiveinformationresources，andappliedwidelyinthefieldofinformationprocessingTherefore，theresearchofautomatictex

3、tcategorizationhasthebroadbusinessprospectsandrealisticsignificanceInthispaperthetraditionalsolutionstosomekeytechnicalproblemsinthefieldofATCarestudiedatfirst，suchasChinesewordsegmentation，featureSelection，featureweight

4、ingandcategorizationalgorithmBycomparingandanalyzingtheimplementationtechnologies，wemakeafurtherdiscussionandputforwardanautomatictextcategorizationprototypebasedonneuralnetworkTheprototypedesignedbasedonmodularizationan

5、dthekeyalgorithmsandfunctionsarepackagedinmodules，SOtheprototypehasagoodportabilityIntheprototype，thecoremodulesarepretreatment，textrepresentationandclassifierIntheprocessofpretreatment，weusedICTCLASwhichhavedevelopedbyt

6、heChineseAcademyofSciencesfortextsegmentationandthenselecttheusefulfeatureforATCNowthemostautomatictextclassificationsystemusedStoplistforselectingfeature，butitisnotsatisfactoryinpracticalapplicationInthispaperweputforwa

7、rdanewmethodtoselectusefulfeatureInthismethod，weselectthefeaturebasedonpartofspeechatfirst，andweeliminatetheotheruselesswordsbyStoplistatlaterWefindthenewmethodimprovestheefficiencyoftheprototypeIntheprocessoftextreprese

8、ntation，weusedifferentmethodsoffeatureselectionandweightingtotakeoutfeaturesfortheconstructionoftextvectorspaceInthismodule，userscanselectthedimensionofvectorspacebytwowaysInthemoduleofclassifierweselectneuralnetworkasou