The Evolution Of Online Public Opinion Among Stakeholders In Public Health Emergencies

Public health emergencies can easily lead to social panic and public opinion crisis.If we can make full use of social media data to discover the evolution trend of online public opinion on public health emergencies,it will help to take effective public opinion measures to guide public opinion in a timely manner.Therefore,this paper conducts topic analysis and sentiment analysis on online public opinion in public health emergencies,so as to obtain the topic evolution and emotion evolution of online public opinion.The main contents of this paper are as follows: First,this paper improves the method of text vectorization.The semantic features of texts obtained by BERT model and the thematic features of texts obtained by LDA model are mixed together.Based on this,this paper designs BERTLDA topic model and BERT-LDA sentiment analysis model.Second,this paper builds classification criteria of lifecycle and stakeholders of public health emergencies.According to the classification criteria,it is divided into 5 lifecycle stages and 11 stakeholders of the COVID-19 epidemic.By text topic clustering,the topics of public opinion of the COVID-19 epidemic,the changes of social concerns throughout the lifecycle,and the topic evolution of the stakeholders are obtained.Third,by text sentiment analysis,the sentiment tendency of public opinion of the COVID-19 epidemic,the changes of social sentiment throughout the lifecycle,and the emotion evolution of the stakeholders are obtained.The conclusions of this paper are as follows: First,BERT-LDA text vector has good classification ability for both topic analysis and sentiment analysis.Second,the social concerns of public health emergencies are relatively consistent in the outbreak period and scattered in other periods.For different stakeholders,stakeholders with the same social role have similar topic evolution patterns.Third,the social sentiment of public health emergencies fluctuates wildly in the contagion period and is relatively stable in other periods.For different stakeholders,stakeholders with the same social role have similar emotion evolution patterns.The innovations of this paper are as follows: First,this paper build BERT-LDA+K-means model to solve topic analysis,and build BERT-LDA+SVM model to solve sentiment analysis.Compared with the traditional topic models and sentiment analysis models,the BERT-LDA model can learn both of semantic information and thematic information of the texts.Second,from the research perspective of lifecycle and stakeholder,this paper conducts topic analysis and sentiment analysis on online public opinion of public health emergencies,so as to summarize the evolution of online public opinion from four aspects: lifecycle stage,stakeholders,public topics and public emotions.Third,this paper summarizes the process of topic evolution,sentiment evolution and "topic-sentiment" evolution of stakeholders,and classifies the similar public opinion evolution patterns among stakeholders.