Classification Of Online Public Opinion During The COVID-19 Period Based On Deep Learnin

Since December 2019,the new type of coronal pneumonia epidemic has started to break out and gradually spread,which has caused countless losses of people’s lives and property.As a result,negative emotions such as fear,worry and panic have emerged in various degrees.In this context,public opinion guidance and maintenance of social stability have become hot topics in social governance.The research on the development of public opinion based on emotion analysis technology is conducive to maintaining social stability and enhancing national confidence in the process of controlling the trend of public opinion.This paper studies and explores the issue of emotion classification based on the data set of 100000 weibo comment texts collected from 230 topics related to the epidemic in weibo.In the data preprocessing stage,this paper uses a variety of data cleaning methods for comment text to de-noise the data.And jieba word segmentation package is used to realize the word segmentation of text sentences,and then the words that do not contain classification information are eliminated using the Harbin Institute of Technology’s stopword list.In the experimental part,this paper uses the Pytorch framework to build a variety of neural networks based on Word2 Vec,Bert and ERNIE pre-training models.The results show that the model based on Bi-LSTM in the model will perform better,and ERNIE’s word embedding effect for Chinese text is the best.Based on the above conclusions,the ERNIE-RCNN(CLS)model proposed in this paper uses ERNIE as a word embedding model to access the Bi-LSTM model,and reduces the dimension of all hidden layer states and word embedding vectors of Bi-LSTM through pooling layer.Finally,the emotion classification is achieved by splicing the results of pooling layer and CLS mapping vector and accessing the full connection layer.The innovation of this paper is that the CLS mapping vector is retained and spliced with the results of the pooling layer.The results show that the ERNIE-RCNN(CLS)model can extract more effective information from more data information.The accuracy of this model in the emotion classification task on the Weibo comment text dataset is 74.95%,and the Macro-F1 value is 0.7163,which is better than the results of other models in this paper.