面向舆情监控的虚假网络信息过滤平台的研究与设计

《面向舆情监控的虚假网络信息过滤平台的研究与设计》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、::10128UDC::20151100093ConvolutionNeuralNetworkCNNword2vecSkip-GramCNNIAbstractWiththedeepeningdegreeofsocialinformation,thenetworkpublicopinionandpeople'sdailylivesarebecominginseparable.Inthefaceofalargenumberofonlinecommunitiesinourcountry,theorganizationswithulteriormotiveshirethenavytofocuson

2、somesociallysensitiveissuesandhotspotstoanonymouslypublishmanyguidingideastoguidethetopicsinthewrongdirection.Thisisnotconducivetosocialstabilityanddevelopment.Therefore,theuseofcomputertechnologytomonitortheInternetpublicopinionhasbecomeahotareaandhasacertainsocialvalue.Thisarticleexploresthefilt

3、eringofdeceptivenetworkinformationinkeytechnologiesofnetworkpublicopinionmonitoringsystem.Asamajorpartofdeceptivenetworkinformation,deceptiveopinionsalsohavecertainharmtosociety.Soitisurgenttoidentifydeceptiveopinions.Thedeceptiveopiniondetectionmainlyextractsopinionscontentcharacteristics,andusem

4、achinelearningtoachievethegoalofdetectingdeceptiveopinion.Mostofthetraditionalmachinelearningisashallowstructure,sothecomplextextcannotberepresented.Atthesametime,deceptiveopinionsdonotonlyreflectintheopiniontextattribute,butalsoreflectinthecommenter'sbehavior.So,fromasinglepointofviewoftextualcon

5、tentattributes,onlyopiniontextattributescannotfullyconsiderthecharacteristicsofdeceptiveopinions,whichcanallleadtolossoffeatures.ThepaperusestheConvolutionNeuralNetwork(CNN)whichistheoneofdepthlearningframeworktoidentifythecontentofdeceptiveopinion.Thepapermarkstheonlinepublichotelopinionstogetala

6、belleddataset.Then,thepaperpreprocessesthedata,suchasChinesewordsegmentationandstopwords,andusesthemodelSkip-Gramoftoolword2vectoobtainthewordvector.IntheaspectofCNNmodelconstruction,thepaperproposesthemodelwhichmixestextcontentattributesandbehavioralattributes.Meanwhile,thetraditionalConvolutionN

7、euralNetworkisbeingimprovedfromthepointofthewordorder,tomakeconvolutionneuralnetworkmoresuitablefortextclassification.Theexperimentalverificationshowsthatthemodelproposedinthispaperhasachievedgoodresultsinthedete