Another Twitter sentiment analysis with Python — Part 11 (CNN + Word2Vec)

《Another Twitter sentiment analysis with Python — Part 11 (CNN + Word2Vec)》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、TheRickest RickyFollowIamadatascientist.Period.Feb23·14minreadAnotherTwittersentimentanalysiswithPython—Part11(CNN+Word2Vec)Thisisthe11thandthelastpartofmyTwittersentimentanalysisproject.Ithasbeenalongjourney,andthroughmanytrialsanderrorsalongtheway,Ihavelearnedcountlessvaluablelessons.Ihaven’t

2、decidedonmynextproject.ButIwilldenitelymaketimetostartanewproject.Youcanndthepreviouspostsfromthebelowlinks.•Part1:Datacleaning•Part2:EDA,Datavisualisation•Part3:Zipf’sLaw,Datavisualisation•Part4:Featureextraction(countvectorizer),N-gram,confusionmatrix•Part5:Featureextraction(Tdfvectorizer)

3、,machinelearningmodelcomparison,lexicalapproach•Part6:Doc2Vec•Part7:Phrasemodeling+Doc2Vec•Part8:Dimensionalityreduction(Chi2,PCA)•Part9:NeuralNetworkswithTdfvectors•Part10:NeuralNetworkswithDoc2Vec/Word2Vec/GloVe*InadditiontoshortcodeblocksIwillattach,youcanndthelinkforthewholeJupyterNoteboo

4、kattheendofthispost.PreparationforConvolutionalNeuralNetworkInthelastpost,Ihaveaggregatedthewordvectorsofeachwordinatweet,eithersummationorcalculatingmeantogetonevectorrepresentationofeachtweet.However,inordertofeedtoaCNN,wehavetonotonlyfeedeachwordvectortothemodel,butalsoinasequencewhichmatche

5、stheoriginaltweet.Forexample,let’ssaywehaveasentenceasbelow.“Ilovecats”Andlet’sassumethatwehavea2-dimensionalvectorrepresentationofeachwordasfollows:I:[0.3,0.5]love:[1.2,0.8]cats:[0.4,1.3]Withtheabovesentence,thedimensionofthevectorwehaveforthewholesentenceis3X2(3:numberofwords,2:numberofvector

6、dimension).Butthereisonemorethingweneedtoconsider.Aneuralnetworkmodelwillexpectallthedatatohavethesamedimension,butincaseofdierentsentences,theywillhavedierentlengths.Thiscanbehandledwithpadding.Let’ssaywehaveoursecondsentenceasbelow.“Ilovedogstoo”withthebelowvectorrepresentationofeachword:I:

7、[0.3,0.5],love:[1.2,0.8],dogs:[0.8,1.2],too:[0.1,0.1]Therstsentencehad3X2dimensionvectors,butthesecondsentencehas4X2dimensionvector.Ourneuralnetworkwon’taccepttheseasinputs.Bypaddingtheinputs,wedecidethemaximumlengthofwordsinasen