Fine-Grained Contextual Predictions for Hard Sentiment Words

《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