[dataguru.cn]Joint Sentiment Topic Model for Sentiment Analysis

《[dataguru.cn]Joint Sentiment Topic Model for Sentiment Analysis》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、JointSentiment/TopicModelforSentimentAnalysisChenghuaLinYulanHeSchoolofEngineering,ComputingandKnowledgeMediaInstituteMathematicsTheOpenUniversityUniversityofExeterMiltonKeynesMK76AA,UKNorthParkRoad,ExeterEX44QF,UKy.l.he.01@cantab.netcl322@exeter.ac.ukABSTRACThavebeenmuchi

2、nterestsinthenaturallanguageprocessingcommunitytodevelopnoveltextminingtechniqueswiththeSentimentanalysisoropinionminingaimstouseautomatedcapabilityofaccuratelyextractingcustomers’opinionsfromtoolstodetectsubjectiveinformationsuchasopinions,at-largevolumesofunstructuredtex

3、tdata.titudes,andfeelingsexpressedintext.Thispaperpro-Amongvariousopinionminingtasks,oneofthemissenti-posesanovelprobabilisticmodelingframeworkbasedonLa-mentclassification,i.e.whetherthesemanticorientationoftentDirichletAllocation(LDA),calledjointsentiment/topicatextisposit

4、ive,negativeorneutral.Whenapplyingma-model(JST),whichdetectssentimentandtopicsimultane-chinelearningtosentimentclassification,mostexistingap-ouslyfromtext.Unlikeothermachinelearningapproachesproachesrelyonsupervisedlearningmodelstrainedfromla-tosentimentclassificationwhichof

5、tenrequirelabeledcor-beledcorporawhereeachdocumenthasbeenlabeledaspos-poraforclassifiertraining,theproposedJSTmodelisfullyitiveornegativepriortotraining.Suchlabeledcorporaareunsupervised.Themodelhasbeenevaluatedonthemovienotalwayseasilyobtainedinpracticalapplications.Also,r

6、eviewdatasettoclassifythereviewsentimentpolarityandsentimentclassificationmodelstrainedononedomainmightminimumpriorinformationhavealsobeenexploredtofur-notworkatallwhenmovingtoanotherdomain.Further-therimprovethesentimentclassificationaccuracy.Prelimi-more,inamorefine-grained

7、sentimentclassificationprob-naryexperimentshaveshownpromisingresultsachievedbylem(e.g.findingusers’opinionsforaparticularproductJST.feature),topic/featuredetectionandsentimentclassificationareoftenperformedinatwo-stagepipelineprocess,byfirstCategoriesandSubjectDescriptorsdetec

8、tingatopic/featureandlaterassigningasentimentla-I.2.7[ArtificialIntelligence]:NaturalLangu