Public Opinion Analysis Algorithm Based On CNN-BILSTM Network And BERT

With the development of computer science,more and more people can use social media easily,such as douyin,weibo and so on.It attracts people to spend a lot of time on the net.There is no limted that people can discuss and share some funny news or interesting things with friends what they get form Internet.Internet public opinion is a comprehensive response of public opinion with strong influence and tendencies.Due to the explosive increase of public opinion messages and the variety of formats,it is necessary to quickly and accurately understand public opinion through the processing and analysis of massive public opinions and build a useful public opinion monitoring system,build a useful public opinion monitoring system,make the accuracy of the case can be more increase,create a safe and green Internet environment.It’s very importent and meaningful for our real life and reasearch.Public opinion analysis mainly includes automatic capture of massive information,topic detection,industry classification,sentiment analysis and other content.In order to improve the accuracy and efficiency of public opinion analysis,the main research of this paper includes the following two aspects:1.The industry classification of the text.Due to the contextual dependence of text structure,CNN is good at extracting local features and cannot get the meaning of the above.Becasuse RNN has no ability to solve the problem of gradient,RNN can not essure the classification result is always right.So we use other ways to avoid those neural networks shortcoming,this paper fuses CNN with BILSTM.After we changed the text into some words,we can use word2 vec to vectorize it.It’s can easliy find form the experimrnts result that because of using this model,it can more full use of CNN’s advantage for local feature extraction while also using BILSTM for bidirectional feature learning to extract global features of text.Then make the features being together usefully,and have a right result by using classification.It’s can be find from the experiments because of this algorithm precison,recall and F1 index,and has a better classification effect.2)Analysis of the sentiment tendency of the text.Because the short text corpus is short and has divergent features,dictionary-based traditional sentiment analysis methods rely heavily on the quality of dictionary construction,and simple deep learning models cannot effectively deal with context-sensitive semantic understanding.In order to better extract text features,this paper uses the BERT-CNN model for public opinion analysis.First,the word vector is obtained through the BERT model,and the high-order features in the sentence and feature dimensionality reduction are further extracted through the CNN network,random Dropout is added to prevent overfitting,after that input feature vector into a fully connected layer,in order to use softmax function have the results of calculated emotion classification.It’s can be seen form the experiments,the way of using Text CNN,Bi Lstm,Bi Lstm+Att and other algorithms can not compare with using the new model,by this way can make the result more right.