资源描述:
《Sentiment classification of movie reviews using contextual valence shifters》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、SentimentClassificationofMovieReviewsUsingContextualValenceShiftersAlistairKennedyandDianaInkpenUniversityofOttawa,Ottawa,ON,K1N6N5,Canada{akennedy,diana}@site.uottawa.caAbstractWepresenttwomethodsfordeterminingthesentimentexpressedbyamoviereview.Thesemanticorientationofarev
2、iewcanbepositive,negative,orneutral.Weexaminetheeffectofvalenceshiftersonclassifyingthereviews.Weexaminethreetypesofvalenceshifters:negations,intensifiersanddiminishers.Negationsareusedtoreversethesemanticpolarityofaparticularterm,whileintensifiersanddiminishersareusedtoincre
3、aseanddecrease,respectively,thedegreetowhichatermispositiveornegative.Thefirstmethodclassifiesreviewsbasedonthenumberofpositiveandnegativetermstheycontain.WeusetheGeneralInquirerinordertoidentifypositiveandnegativeterms,aswellasnegationterms,intensifiers,anddiminishers.Wealsou
4、sepositiveandnegativetermsfromothersources,includingadictionaryofsynonymdifferencesandaverylargeWebcorpus.Tocomputecorpus-basedsemanticorientationvaluesofterms,weusetheirassociationscoreswithasmallgroupofpositiveandnegativeterms.Weshowthatextendingtheterm-countingmethodwith
5、contextualvalenceshiftersimprovestheaccuracyoftheclassification.ThesecondmethodusesaMachineLearningalgorithm,SupportVectorMachines.Westartwithunigramfeaturesandthenaddbigramsthatconsistofavalenceshifterandanotherword.Theaccuracyofclassificationisveryhigh,andthevalenceshifterb
6、igramsslightlyimproveit.Thefeaturesthatcontributetothehighaccuracyarethewordsinthelistsofpositiveandnegativeterms.Previousworkfocusedoneithertheterm-countingmethodortheMachineLearningmethod.Weshowthatcombiningthetwomethodsachievesbetterresultsthaneithermethodalone.Keywords:
7、Sentimentclassification,semanticorientation,valenceshifters,machinelearning,evaluation.11IntroductionDocumentscanbecategorizedinvariousways,forexamplebysubject,genre,orthesentimentexpressedinthedocument.Wefocusonsentimentclassification(intopositiveornegativeopinions).Oneusefu
8、lapplicationofsentimentclassificationisinquestionanswering.Caseswhereauserisaskinga