Sentiment Analysis in Twitter Based on Knowledge Graph and ...

In this proposal, tweets are represented as graphs; then, graph similarity metrics and a Deep Learning classification algorithm are applied to produce sentiment ... Electronics8monthsagoSentimentAnalysisinTwitterBasedonKnowledgeGraphandDeepLearningClassificationFernandoAndresLovera,YudithCoromotoCardinale,MasunNabhanHomsiThetraditionalwaytoaddresstheproblemofsentimentclassificationisbasedonmachinelearningtechniques;however,thesemodelsarenotabletograspalltherichnessofthetextthatcomesfromdifferentsocialmedia,personalwebpages,blogs,etc.,ignoringthesemanticofthetext.Knowledgegraphsgiveawaytoextractstructuredknowledgefromimagesandtextsinordertofacilitatetheirsemanticanalysis.ThisworkproposesanewhybridapproachforSentimentAnalysisbasedonKnowledgeGraphsandDeepLearningtechniquestoidentifythesentimentpolarity(positiveornegative)inshortdocuments,suchaspostsonTwitter.Inthisproposal,tweetsarerepresentedasgraphs;then,graphsimilaritymetricsandaDeepLearningclassificationalgorithmareappliedtoproducesentimentpredictions.Thisapproachfacilitatesthetraceabilityandinterpretabilityoftheclassificationresults,thankstotheintegrationoftheLocalInterpretableModel-agnosticExplanations(LIME)modelattheendofthepipeline.LIMEallowsraisingtrustinpredictivemodels,sincethemodelisnotablackboxanymore.Uncoveringtheblackboxallowsunderstandingandinterpretinghowthenetworkcoulddistinguishbetweensentimentpolarities.Eachphaseoftheproposedapproachconformedbypre-processing,graphconstruction,dimensionalityreduction,graphsimilarity,sentimentprediction,andinterpretabilitystepsisdescribed.Theproposaliscomparedwithcharactern-gramembeddings-basedDeepLearningmodelstoperformSentimentAnalysis.Resultsshowthattheproposalisabletooutperformsclassicaln-grammodels,witharecallupto89%andF1-scoreof88%.PublisherURL:https://www.mdpi.com/2079-9292/10/22/2739DOI:10.3390/electronics10222739YoumightalsolikeDiscover&DiscussImportantResearchKeepingup-to-datewithresearchcanfeelimpossible,withpapersbeingpublishedfasterthanyou'lleverbeabletoreadthem.That'swhereResearchercomesin:we'resimplifyingdiscoveryandmakingimportantdiscussionshappen.Withover19,000sources,includingpeer-reviewedjournals,preprints,blogs,universities,podcastsandLiveeventsacross10researchareas,you'llnevermisswhat'simportanttoyou.It'slikesocialmedia,butbetter.Oh,andweshouldmention-it'sfree.Researcherdisplayspubliclyavailableabstractsanddoesn’thostanyfullarticlecontent.Ifthecontentisopenaccess,wewilldirectclicksfromtheabstractstothepublisherwebsiteanddisplaythePDFcopyonourplatform.Clickstoviewthefulltextwillbedirectedtothepublisherwebsite,whereonlyuserswithsubscriptionsoraccessthroughtheirinstitutionareabletoviewthefullarticle.