基于优化bp神经网络的微博舆情预测模型-研究

《基于优化bp神经网络的微博舆情预测模型-研究》由会员上传分享，免费在线阅读，更多相关内容在教育资源-天天文库。

1、AbstractWiththerapiddevelopmentofIntemettechnology,micro—bloghasbecomeanimportantpartofpeople’Slives．Publicopinioncausedbythemicro—blogattractsmoreandmorepeople’Sattention．Sincethemicro—bloginformation’Spropagationvelocityisquickly,spreadwidelyandarbitrary

2、natureofmicro—blog，SOtheinformationonthemicro-bloghastrueandfalse．Positiveandnegativemicro—blogpublicopinionwillhavedifferenteffectsonpeople’Slives，someofthenegativemicro—blogpublicopinionwillevenconstituteacrisis，thenaseriousimpactonpublicsafety．Therefore

3、，thestudyonpredictedmicro—blogpublicopinionhasapracticalsignificance．Predictedmicro—blogpublicopinion，wemustgetdatawhichCanbeexpressedmicro—blogpublicopinionfirstly．Thepaperusesthediscretetimeseriestodescribemicro—blogpublicopinion’Strends．Inthispaper,weus

4、eSinamicro-blogplatformforbackground，accordingthehotmicro—blogtopictextextraction，analysis，forecastmicro—blogpublicopinion．Gettimeseriesstep：one诵tllSinamicro—blogAPIinterface，accesstomicro—blogtogetmicro—blogtextinsometime；Second，accordingthecharacteristic

5、softhecorrespondingpretreatmentmicro—blogtext，usethemethodsofstatisticalmicro-blogtopicandfoundthatmicro—bloghottopic；Third，statisticalthenumberofrepliesandforwardingnumberofmicro—bloghottopicforsometime，andusethenumbertoconsistofpublicopinionpredictionmod

6、el’Sexperimentaldata．BPneuralnetworkCanbebetterfitthenonlinearvariationofmicro—blogpublicopinion’Stimeseries，whichcanbeusedtopredictthemicro—blogpublicopinion，buttherearesomelimitations：BPneuralnetwork’Slearningalgorithmhastheweaknessofforgettingthealready

7、learningsamples．Whenthereisnoiseinthesample，theremaycausepoorperformanceonB．Pneuralnetwork；BPneuralnetworkalsohasaslowspeedofconvergence，andeasytofallintolocalminima．Wedidtwotasks：First，changethenetwork’SstructuretoimproveBPneuralnetwork．BehindtheBPneuraln

8、etworkinputlayerneurons，weaddalayertostoretheinputlayer’Shistorydata．Whenthesampleshaveanoisedata，thelayercandelaynetworkparameterstoimprovetheperformanceofBPneuralnetwork．ThesecondistouseGSAtooptimizethenetw