卷积神经网络相关外文文献翻译中英文.docx
- 文档编号:14449775
- 上传时间:2023-06-23
- 格式:DOCX
- 页数:15
- 大小:29.11KB
卷积神经网络相关外文文献翻译中英文.docx
《卷积神经网络相关外文文献翻译中英文.docx》由会员分享,可在线阅读,更多相关《卷积神经网络相关外文文献翻译中英文.docx(15页珍藏版)》请在冰点文库上搜索。
卷积神经网络相关外文文献翻译中英文
卷积神经网络相关外文文献翻译中英文2020
英文
Socialmediasentimentanalysisthroughparalleldilatedconvolutionalneuralnetworkforsmartcityapplications
MuhammadAlam,FazeelAbid,etc
Abstract
DeepLearningisconsideredtoleveragesmartcitiesthroughsocialmediasentimentanalysis.Thedigitalcontentinsocialmediacanbeusedformanysmartcityapplications(SCAs).Classicalconvolutionalneuralnetworks(CNNs)arechallengingtoparallelizeandinsufficienttocapturelongtermcontextualsemanticfeaturesforsentimentanalysis.Inthisperspective,thispaperinitiallyproposesadomain-specificdistributedwordrepresentation(DS-DWR) withaconsiderablysmallcorpussizeinducedfromtextualresourcesinsocialmedia.InDS-DWR,differentDistributedWordRepresentationsareconcatenatedtobuildsrichrepresentationsovertheinputsequence,whichisworthwhileforinfrequentandunseenterms.Second,adilatedconvolutionalneuralnetwork(D-CNN) ,whichiscomposedofthreeparalleldilatedconvolutionalneuralnetwork(PD-CNN) layersandaglobalaveragepooling(GAP) layer.Ourconsideredparalleldilatedconvolutionreducesdimensionandincorporatesanextensioninthesizeofreceptivefieldswithoutthelossoflocalinformation.Further,thelong-termcontextualsemanticinformationisachievedbytheuseofdifferentdilationrates.Experimentsdemonstratethatourarchitectureaccomplishescomparableresultswithmultiplehyperparameterstuningforbetterparallelismwhichleadstotheminimizedcomputationalcost.
Keywords:
Socialmediasentimentanalysis,Smartcityapplications,Paralleldilatedconvolutionalneuralnetwork(PD-CNN),Domain-specificdistributedwordrepresentation(DS-DWR)
Introduction
Thedevelopmentofsmartcityapplications(SCAs)inthelastdecadehasshownflexibledesignandvaluableservices.SmartCitiesareacombinationoftechnicalandsocialadvancements,e.g., InternetaccessonMobile,SocialDataAnalytics,andInternetofThings(IoT) [1].Theseadvancementscanfacilitatetomanageresourcesproductivelyprimarilythroughsocialnetworks.OurstudyendorsesanovelwayforthesmartcitiestowardsthebuildingofintelligentandadaptableSCAs.Thesocialmediahasbecomeapotentialinformationsourceforsocialinformationminingtodominatepeople’sopinions.Inourperspective,userscanbeviewedassocialsensors,andopinionsareresponsesignalsasitisreal-timecharacteristicsofsocialdata.Theseopinionsareextensiveonsocialmediaconsistofshortsentencesandcontainusefulinformationinnumerouscontexts.Therefore,socialmediadataassocialsensorscanbeemployedtodiscovervaluableinformation.Adevelopingpatternintheareaofsocialdataminingistoconcentrateonsocialtextualdatahavingopinionsinsteadofconcentratingonextensivenumericalanalysis.Socialmediasentimentanalysisisagrowingtechniquetocomprehendtheopinionsofindividualsthroughsocialnetworks.WhileSCAscanutilizethebenefitsofsocialmediasentimentanalysisastheopinionscanberelatedtoanyeventorasourceofestimatingpreferences,dislikes,andpatterns.
SCAscanactas“activators”byfocusingonimprovementsintechnological,political,andorganizationalaspectsthroughsocialmedia.Also,itcanenhancepeople’sattitudesandempowerthemtobuildupafeasibleenvironmentinSmartcities [2].Socialnetworksaretheprimarysourceofopinionsandevents;however,themainissueissubstantialsocialcontentthatrequiressmartapproachestofilteroutnoisydata.Theprerequisiteofnoisefilteringofsocialmediadatainvolvesautomaticclassificationtechniquesforvaluablefacts.Italsosuffersafewdifficulties;forexample,sentencesareshortandcontainabbreviationsaswellasspellingmistakes.Thesemanticanalysiscanbeusedtomanageabundantsocialmediadatatoinducesemanticforimprovementsintermsofintegrationandreusability [3].Similarly,socialmediaminingtechniquescanutilizetocollectandprocesssocialdatapostsonFacebook,Twitter,andInstagramforSCAs [4].WechooseTwitter,whichhasaconsiderablenumberofpostsandconsistsoflessthan140characters [5].Thissocialmediadatacanbeusedforsocialmediasentimentanalysisandconsideredappropriatetoexaminepeople’sopinionsofsmartcityservices.
Currently,MachineLearning(ML)methodologiesutilizedIoTandBigDatafortheenhancementsofsmartcityservices [6].AmongmanyMLmethodsonclassifyingtext,NaıveBayes(NB)isutilizedforspamdetection [7],topicdetection [8],sentimentanalysis [9],recommendationsystems [10].Inasocialcontext,populartechniquessupportvectormachine(SVM)hasbeenutilizedtocharacterizetweetstogetthetrends [11].However,therearesomeissuesrelatedtotheextractionofthefeatureswithvariable-sizesequencescomposition.Whereasasubfield,deepLearning(DL),favorablyprovidesintelligentservices.DLemploysartificialneuralnetworks(ANNs)architecturesinordertolearnhigh-levelfeaturesforsocialdatawithmultiplelayers.ThesefeaturesareefficientenoughtoreplacemanyMLmethodologies,especiallywhenaddressingsocialmediadataforpredictions.Further,thesemethodologiesareprogressivelyappliedtosupervisedandunsupervisedclassificationproblems [12], [13], [14].
Deeplearning(DL)persuadesresearchersbyusingANNsthatempowersefficientlearningoffeatureswithoutrequiringcomplexfeature-engineering [15], [16].DLmethodologiesworkbyexecutingthefeatureextractionandclassificationtaskthroughinitiatingtherepresentationofasequenceofwordsviamultiplyingwithassociatedweightmatrixasaone-hotvector [17].Thissequenceofeachwordisformulatedbycontinuousvectorspaceinputtedtoaneuralnetworktoprocessthesuccessionofwordswiththeassistanceofvariouslayersforpredication.Thisprocessinghasanimpactonthetrainingsettowardsincreasingtheclassificationaccuracy,asexplainedin [18].Convolutionalneuralnetwork(CNN)hasaccomplishedexcellentoutcomesofsentence-levelclassification [19], [20].ItcanlearnfromdifferentdistributedwordrepresentationssuchasWord2Vec [21],FastText [22] andGloVe [23] byprojectingthewordsintolowerdimensionswithdensespacevector.AtechniqueofCNNtoextractthefeaturesofthesentenceandjoinedthesefeatureswithhandcraft-featuresfortherelationclassificationproposedin [24].However,traditionalCNNsareunabletoholdlongtermsemanticfeatures.Comparatively,auniquedesignofCNNistheutilizationofdilatedconvolution.Iteliminatestheimplicationsofinformationloss,whichisduetoconventionaldown-samplingapproachesintraditionalpoolingoperationsandlikewiseinthestrideconvolution.Further,itcanexpandthesizeofreceptivefieldsattheexponentiallevelwithoutconcerningadditionalparameters.Thus,itbecomesfeasibleforthedilatedconvolutiontocapturelongtermsemanticinformation.
AlthoughtherearenumerousSCAs,stillitappearsasacollectionoflooselycoupledrelationwithsocialmedia,whichisneedtobestrengthened.ThecurrentresearchisutilizingadeeplearningmethodologythatautomaticallyidentifiesinformationonsocialsensorsbyconsideringaparalleldilatedconvolutionalneuralNetwork(PD-CNN).Fromthebestofourreview,PD-CNNiscapableofpredictingappropriateinformationasitlearnsfeaturesfromDS-DWRsusingdifferentdistributedwordrepresentations,asmentionedabove.BeyondtheimprovementinSCAs,severalfavorablehyper-parameterswhileutilizingDLforsocialmediasentimentanalysisareconsidered.Ourworknotonlyintroducesthenewwayofsocialmediasentimentanalysisbutalsousefultomodernizesmartcities.Thefollowingarethecontributionsofthiswork:
•Thegenerationofdomain-specificdistributedwordrepresentationisexploredtodiscoverthefeaturesfortheenhancedexhibitionoftheclassifierincomparisonwithstandardmethods.
•Theparallelisminthedilatedconvolutionalneuralnetworkcontainsthreedilatedconvolutionlayerscontainingdifferentdilationrateswithglobalaveragepoolinginanovelmanner.
•Lastly,withoutthecomplexfeature-engineering,theworkproceedswiththeemploymentoftheDLmethodtoprovideintelligentSmartcityapplicationsthroughsocialmediasentimentanalysisbasedonthesocialsensor(Tweets)toaugmentthepeople’sperceptioninsmartcities.
Relatedwork
SmartCityandsocialmediaanalyticscontinuousgrowthcontributedtoanadvancedlevelintechnologicalandurbanadvancements.Onlyafewstudiesinthelastyearsclarifythecomplementarycharacteristicsofthesocial-technicalecosystem.However,theirmutualsignificanceisnotyetwhollyidealized [25].Theworksonsustainabilitywhichexplainthesmartcitiesbyrelatinginnovationstosocialnetworksexplainedin [26], [27].Socialusersofsmartcityservicesarewellinformedabouttheavailabilityandproductivityoftheseservices,asdescribedin [28].ItImpliestherequirementformorepracticalityinsmartcitesresearchregardingsocialnetworks.Forthesustainabilityinsmartcitiesresearch,onehastoconcentrateonthesocialcontextrelatedtosmartcities [29].InthisWay,smartcityapplications(SCAs)prompttheintegrationofsocialnetworksandsmartcitiesserviceshavingopinionsandprospects,asexplainedin [30], [31].Thus,socialnetworkshaveaprofoundimpactonSCAs.Socialnetworkscontaininformalmethodsofinteractionsandopinions.Weattempttoanalyzehowtheseopinionscontributetopromotethementionedservices.Further,thesenetworksallowingtosharestatusupdatemessages.Thesemessageshaveadditional(meta)information,e.g., name,time,location,andhashtags.It
- 配套讲稿:
如PPT文件的首页显示word图标,表示该PPT已包含配套word讲稿。双击word图标可打开word文档。
- 特殊限制:
部分文档作品中含有的国旗、国徽等图片,仅作为作品整体效果示例展示,禁止商用。设计者仅对作品中独创性部分享有著作权。
- 关 键 词:
- 卷积 神经网络 相关 外文 文献 翻译 中英文