情感分析技术在美团的探索与应用 2021年5月

2021年5月，美团NLP中心开源了迄今规模最大的基于真实场景的中文属性级情感分析数据集ASAP ，该数据集相关论文被自然语言处理顶会NAACL2021录用，同时该数据集加入中文开源数据计划千言，将与其他开源数据集一起推动中文信息处理技术的进步。本文回顾了美团情感分析技术的演进和在典型业务场景中的应用，包括篇章/句子级情感分析、属性级情感分析和观点三元组分析。在业务应用上，依托情感分析技术能力构建了在线实时预测服务和离线批量预测服务。截至目前，情感分析服务已经为美团内部十多个业务场景提供了服务。

文章图片

文章图片

文章图片
【情感分析技术在美团的探索与应用】参考文献
[1]https://github.com/Meituan-Dianping/asap.[2]BuJ,RenL,ZhengS,etal.ASAP:AChineseReviewDatasetTowardsAspectCategorySentimentAnalysisandRatingPrediction.InProceedingsofthe2021ConferenceoftheNorthAmericanChapteroftheAssociationforComputationalLinguistics:HumanLanguageTechnologies.2021.[3]https://www.luge.ai/[4]Zhang,L.,S.Wang,andB.Liu.''DeepLearningforSentimentAnalysis:ASurvey.''WileyInterdisciplinaryReviews:DataMiningandKnowledgeDiscovery(2018):e1253.[5]Liu,Bing.''Sentimentanalysisandopinionmining.''Synthesislecturesonhumanlanguagetechnologies5.1(2012):1-167.[6]Peng,Haiyun,etal.''Knowingwhat,howandwhy:Anearcompletesolutionforaspect-basedsentimentanalysis.''InProceedingsoftheAAAIConferenceonArtificialIntelligence.Vol.34.No.05.2020.[7]Zhang,Chen,etal.''AMulti-taskLearningFrameworkforOpinionTripletExtraction.''InProceedingsofthe2020ConferenceonEmpiricalMethodsinNaturalLanguageProcessing:Findings.2020.[8]YoonKim.2014.Convolutionalneuralnetworksforsentenceclassification.arXivpreprintarXiv:1408.5882.[9]PengZhou,WeiShi,JunTian,ZhenyuQi,BingchenLi,HongweiHao,andBoXu.2016.Attention-basedbidirectionallongshort-termmemorynetworksforrelationclassification.InProceedingsofthe54thAnnualMeetingoftheAssociationforComputationalLinguistics(Volume2:ShortPapers),pages207–212.[10]Devlin,Jacob,etal.“Bert:Pre-trainingofdeepbidirectionaltransformersforlanguageunderstanding.”arXivpreprintarXiv:1810.04805(2018).[11]杨扬、佳昊等.美团BERT的探索和实践.[12]Pontiki,Maria,etal.''Semeval-2016task5:Aspectbasedsentimentanalysis.''Internationalworkshoponsemanticevaluation.2016.[13]Pontiki,M.,etal.''SemEval-2014Task4:AspectBasedSentimentAnalysis.''InProceedingsofInternationalWorkshoponSemanticEvaluationat(2014).[14]YequanWang,MinlieHuang,andLiZhao.2016.Attention-basedlstmforaspect-levelsentimentclassification.InProceedingsofthe2016ConferenceonEmpiricalMethodsinNaturalLanguageProcessing,pages606–615.[15]SaraSabour,NicholasFrosst,andGeoffreyEHinton.2017.Dynamicroutingbetweencapsules.InAdvancesinneuralinformationprocessingsystems,pages3856–3866.[16]ChiSun,LuyaoHuang,andXipengQiu.2019.Utilizingbertforaspect-basedsentimentanalysisviaconstructingauxiliarysentence.arXivpreprintarXiv:1903.09588.[17]QingnanJiang,LeiChen,RuifengXu,XiangAo,andMinYang.2019.Achallengedatasetandeffectivemodelsforaspect-basedsentimentanalysis.InProceedingsofthe2019ConferenceonEmpiricalMethodsinNaturalLanguageProcessingandthe9thInternationalJointConferenceonNaturalLanguageProcessing(EMNLP-IJCNLP),pages6281–6286.[18]Wu,Zhen,etal.''GridTaggingSchemeforEnd-to-EndFine-grainedOpinionExtraction.''InProceedingsofthe2020ConferenceonEmpiricalMethodsinNaturalLanguageProcessing:Findings.2020.[19]Liu,Yinhan,etal.''Roberta:Arobustlyoptimizedbertpretrainingapproach.''arXivpreprintarXiv:1907.11692(2019).[20]Clark,Kevin,etal.''Electra:Pre-trainingtextencodersasdiscriminatorsratherthangenerators.''arXivpreprintarXiv:2003.10555(2020).0-[21]TimothyDozatandChristopherD.Manning.2017.Deepbiaffineattentionforneuraldependencyparsing.In5thInternationalConferenceonLearningRepresentations,ICLR2017.作者介绍