基于類間距離最大化的特征抽取方法研究.pdf

1、獨創(chuàng)性聲明本人聲明所呈交的學(xué)位論文是本人在導(dǎo)師指導(dǎo)下進(jìn)行的研究工作及取得的研究成果。據(jù)我所知，除了文中特別加以標(biāo)注和致謝的地方外，論文中不包含其他人已經(jīng)發(fā)表或撰寫過的研究成果，也不包含為獲得或其他教育機(jī)構(gòu)的學(xué)位或證書而使用過的材料。與我一同工作的同志對本研究所做的任何貢獻(xiàn)均已在論文中作了明確的說明并表示謝意。學(xué)位論文作者簽名：簽字日期：年月日學(xué)位論文版權(quán)使用授權(quán)書本學(xué)位論文作者完全了解江西師范大學(xué)研究生院有關(guān)保留、使用學(xué)位論文的規(guī)定，有

2、權(quán)保留并向國家有關(guān)部門或機(jī)構(gòu)送交論文的復(fù)印件和磁盤，允許論文被查閱和借閱。本人授權(quán)江西師范大學(xué)研究生院可以將學(xué)位論文的全部或部分內(nèi)容編入有關(guān)數(shù)據(jù)庫進(jìn)行檢索，可以采用影印、縮印或掃描等復(fù)制手段保存、匯編學(xué)位論文。(保密的學(xué)位論文在解密后適用本授權(quán)書)學(xué)位論文作者簽名：簽字日期：年月日導(dǎo)師簽名：簽字日期：年月日ABSTRACTWiththepopularizationoftheIntemetandextensiveapplicationof

3、theonlineelectronictexts，alongwiththeexplosivegrowthofthetextsTextclassification，animportanttechnologytoorganizeandmanagelargeamountsoftextinformation，hasbeenwidelyusedinmanyfields，whichhasbecameahottopicindataminingarea

4、Howeverthehighdimensionaloftextdataisanimportantproblemtopersecutetheperformanceoftextclassification。ThehugenumberoffeaturespacenotoI坶increasesthecomputationalcomplexitybutalsoaffectstheclassificationperformanceandgenera

5、lizationability，resulting”overlearn”phenomenonTherefore，dimensionreductionhasbecomeanessentialstepintextclassificationGenerallyspeaking，featureselectionandfeatureextractionaretwocommonmethodsteducefeaturedimensionFeature

6、selectionconstructsaratingfunctiontoscorethefeatures，thenselectsfeaturesaccordingthemtiIlgsequence，whichisasimpleandstraightforwardmethodBycomparison，featureextractiongeneratesanewsetoffeaturesbyassemblingtheoriginalfeat

7、ures，thenselectsthenewfeatures，whichcansolvetheproblemofsynonymsandpolysemousproblemsHowever，duetothefeatureselectionneglectsthesynonymsandpolysemousphenomenon，resultingthelowperformanceoftextclassificationTherefore，this

8、paperpresentsafeatureextractionbasedmethodwhichmaximizesthedistancebetweenclassesandconstructsalloptimalfunctionbymaximizingthedistancebetweendifferentcategoriesofdocumentstoobtainmappingmatrixEvaluationonChinesetextsrel

9、easedbyFudanUniversityshowsthattheeffectivenessofOUl“modelItalsoshowsthatourmodelsignificantlyoutperformschisquarestatisticbasedmodelcurrentlyintheresearchfieldKeyWord：textclassification；dimensionreduction；featureextract