基于Winnow算法的反垃圾郵件引擎的設(shè)計(jì)與實(shí)現(xiàn).pdf

1、東南大學(xué)碩士學(xué)位論文基于Winnow算法的反垃圾郵件引擎的設(shè)計(jì)與實(shí)現(xiàn)姓名：張麗申請(qǐng)學(xué)位級(jí)別：碩士專業(yè)：控制理論與控制工程指導(dǎo)教師：黃東20060301竺!里墜堅(jiān)ABSTRACTEmailhasbecomeanimportantmeansofcommunicationandanimportantcomponentoftheenterpriseoperation，howeverasabyproductofEmail，spamisincrea

2、singlyaffectingpeople‘sdailylives，andanti—spamhasbecomeamajorglobalresearchtopicAmongthevarioustechniquestosolvespamproblems，beginwiththemailcontent，andusetextcategorizationtechnique，thenstudyandconstructclassificationon

3、trainingsets，atlasttestthesystemperformanceusingtestsetsisacommonlyusedmethodRecentlythespamcenterbeginstomovetoChina，sotheresearchofChineseantispamisimportantThisdissertationfocusonChinesespamVarioustechniquesrequiredin

4、antispamenginewerediscussedandthesystemmoduleweredesigned，includingpreprocessmodule，trainingmodule，classifymoduleandfeedbackmoduleandtheimplementationofpartimportantmodulesweregivenAtlastthesystemofChineseanti—spamengine

5、basedonWinnowalgorithmwasrealizedSpecifically，thisarticleincludesthefollowingmainparts：1)PreprocessmoduleincludemaildecodeandChinesewordsegmentationAdetailedBase64andQPencodingstandardsanddecodealgorithmsweregiveninthema

6、ildecodemoduleandinordertobettermaintainingthedictionary，animprovedfullbinarysearchmaximalmatchalgorithmforChinesewordsegmentationwasadoptedintheChinesewordsegmentationmodule2)Winnowalgorithmwasadoptedtoconstructtheclass

7、ificationinthetrainingmoduleThetrainingofWinnowisonlineandm／stakedrivenFurthermore。WinnowissuitableforfeedbackBothpositiveWinnowandbalancedWinnowwererealizedinthissystemandaftertestbalancedWinnowwasfoundsuperiortopositiv

8、eWinnow3】Inclassifymodule，itwasfoundthattheoriginalmethodtosetthresholdvaluewouldleadtotoolowRecall，afterthethresholdadjusted，theeffectswouldbesignificantincreaseinsystem，atlastthewaytoadjustthresholdvaluewasgivenKeyword