资源描述：
《2017-PAKDD-Dependency-tree Based Convolutional Neural Networks for Aspect Term Extraction》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、Dependency-treeBasedConvolutionalNeuralNetworksforAspectTermExtractionHaiYey,ZichaoYany,ZhunchenLuoz??andWenhanChaoyySchoolofComputerScienceandEngineering,BeihangUniversity,Beijing,ChinazChinaDefenseScienceandTechnologyInformationCenter,Beijing,Chinafyehai,yanzichao,chaowenhang@buaa.edu.cn,zhunchen
2、luo@gmail.comAbstract.Aspecttermextractionisoneofthefundamentalsubtasksinaspect-basedsentimentanalysis.Previousworkhasshownthatsen-tences'dependencyinformationiscriticalandhasbeenwidelyusedforopinionmining.Withrecentsuccessofdeeplearninginnaturallanguageprocessing(NLP),recurrentneuralnetwork(RNN)ha
3、sbeenproposedforaspecttermextractionandshowsthesuperiorityoverfeature-richCRFsbasedmodels.However,becauseRNNisasequentialmodel,itcannote ectivelycapturetree-baseddependencyinformationofsentencesthuslimitingitspracticability.Inordertoe ectivelyexploitsentences'dependencyinformationandleveragethee ec
4、tivenessofdeeplearning,weproposeanoveldependency-treebasedconvolutionalstackedneuralnetwork(DTBCSNN)foraspecttermextraction,inwhichtree-basedconvolutionisintroducedoversentences'dependencyparsetreestocap-turesyntacticfeatures.Ourmodelisanend-to-enddeeplearningbasedmodelanditdoesnotneedanyhuman-craf
5、tedfeatures.Furthermore,ourmodelis
exibletoincorporateextralinguisticfeaturestofurtherboostthemodelperformance.Tosubstantiate,resultsfromexperimentsonSemEval2014Task4datasets(reviewsonrestaurantandlaptopdomain)showthatourmodelachievesoutstandingperformanceandoutperformstheRNNandCRFbaselines.Keyword
6、s:Aspecttermextraction,Dependencyinformation,Tree-basedconvolution,Deeplearning1IntroductionAspect-basedsentimentanalysis(oropinionmining)aimstoidentifytheopin-ionsinagivendocument.Toachievethisgoal,sixsubtasksshouldbeconsideredandaspecttermextractionisoneoftheimportantsubtasks[1].Aspecttermsareatt
7、ributes(orproperties)oftheentitythatopinionexpresseson.Forexam-ple,giventheproductreviewIlovethewaytheentiresuiteofsoftwareworkstogether",theaspecttermissuiteofsoftware".??Correspondingauthor.2HaiYe,Zicha
