资源描述：
《Text classification from labeled and unlabeled documents using EM》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。
1、MachineLearning,,1{34()cKluwerAcademicPublishers,Boston.ManufacturedinTheNetherlands.TextClassi cationfromLabeledandUnlabeledDocumentsusingEMyKAMALNIGAMknigam@cs.cmu.eduzyANDREWKACHITESMCCALLUMmccallum@justresearch.comySEBASTIANTHRUNthrun@cs.cmu.eduyTOMMITCHELLtom.mitchell@cmu.eduySchoolofComputerSc
2、ience,CarnegieMellonUniversity,Pittsburgh,PA15213zJustResearch,4616HenryStreet,Pittsburgh,PA15213ReceivedMarch15,1998;RevisedFebruary20,1999Editor:WilliamW.CohenAbstract.Thispapershowsthattheaccuracyoflearnedtextclassi erscanbeimprovedbyaugmentingasmallnumberoflabeledtrainingdocumentswithalargepoolo
3、funlabeleddocu-ments.Thisisimportantbecauseinmanytextclassi cationproblemsobtainingtraininglabelsisexpensive,whilelargequantitiesofunlabeleddocumentsarereadilyavailable.WeintroduceanalgorithmforlearningfromlabeledandunlabeleddocumentsbasedonthecombinationofExpectation-Maximization(EM)andanaiveBayesc
4、lassi er.Thealgorithm rsttrainsaclassi erusingtheavailablelabeleddocuments,andprobabilisticallylabelstheunlabeleddocuments.Itthentrainsanewclassi erusingthelabelsforallthedocuments,anditeratestoconvergence.ThisbasicEMprocedureworkswellwhenthedataconformtothegenerativeassumptionsofthemodel.Howeverthe
5、seassumptionsareoftenviolatedinpractice,andpoorperformancecanresult.Wepresenttwoextensionstothealgorithmthatimproveclassi cationaccuracyundertheseconditions:(1)aweightingfactortomodulatethecontributionoftheunlabeleddata,and(2)theuseofmultiplemixturecomponentsperclass.Experimentalresults,obtainedusin
6、gtextfromthreedi erentreal-worldtasks,showthattheuseofunlabeleddatareducesclassi cationerrorbyupto30%.Keywords:textclassi cation,Expectation-Maximization,integratingsupervisedandunsuper-visedlearning,combininglabeledandunlabeleddata,Bayesianlearning1.IntroductionConsidertheproblemofautomaticallyclas
7、sifyingtextdocuments.Thisproblemisofgreatpracticalimportancegiventhemassivevolumeofonlinetextavail-ablethroughtheWorldWideWeb,Internetnewsfeeds,electronicmail,corporatedatabases,medicalpatientrecordsa
