欢迎来到天天文库
浏览记录
ID:40715874
大小:142.95 KB
页数:6页
时间:2019-08-06
《Fine-Grained Contextual Predictions for Hard Sentiment Words 》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、Fine-GrainedContextualPredictionsforHardSentimentWordsSebastianEbertandHinrichSchutze¨CenterforInformationandLanguageProcessingUniversityofMunich,Germanyebert@cis.lmu.de,inquiries@cislmu.orgAbstractsentimentanalysis:closelyrelatedsenseswiththesamesentimentshou
2、ldbemergedwhereassubtleWeputforwardthehypothesisthathigh-semanticdistinctionsthatgiverisetodifferentpo-accuracysentimentanalysisisonlypos-laritiesshouldbedistinguished.sibleifwordsenseswithdifferentpolar-Thedatastructurein(iii)isastatisticalclassi-ityareaccura
3、telyrecognized.Wepro-ficationmodelinthesimplestcase.Wewillgivevideevidenceforthishypothesisinacaseoneotherexamplefor(iii)below:itcanalsobeastudyfortheadjective“hard”andproposesetofcentroidsofcontextvectorrepresentations,contextuallyenhancedsentimentlexiconswith
4、amappingofthesecentroidstothesenses.thatcontaintheinformationnecessaryforIfsentiment-relevantsensedisambiguationissentiment-relevantsensedisambiguation.thefirststepinsentimentanalysis,thenpower-Anexperimentalevaluationdemonstratesfulcontextualfeaturesarenecessa
5、rytosupportthatsenseswithdifferentpolaritycanbemakingfine-graineddistinctions.Ourthirdcon-distinguishedwellusingacombinationoftributionisthatweexperimentwithdeeplearn-standardandnovelfeatures.ingasasourceofsuchfeatures.Welookattwotypesofdeeplearningfeatures:wor
6、dem-1Introductionbeddingsandneuralnetworklanguagemodelpre-dictions(Section4).Weshowthatdeeplearn-Thispaperdealswithfine-grainedsentimentanal-ingfeaturessignificantlyimprovetheaccuracyysis.Weaimtomakethreecontributions.First,ofcontext-dependentpolarityclassificati
7、on(Sec-basedonadetailedlinguisticanalysisofcontextstion5).oftheword“hard”(Section3),wegiveevidencethathighlyaccuratesentimentanalysisisonlypos-2Relatedworksibleifsenseswithdifferentpolarityareaccu-ratelyrecognized.InitialworkonsentimentanalysiswaseitherbasedSe
8、cond,basedonthisanalysis,weproposetoonsentimentlexiconsthatlistedwordsasposi-returntoalexicon-basedapproachtosentimenttiveornegativesentimentindicators(e.g.,Turneyanalysisthatsuppo
此文档下载收益归作者所有