Research On The Evolution Of Public Opinion Of COVID-19 Based On Microblog Visualization

With the rapid development of the mobile Internet,social media such as Weibo has become the main platform for people to discuss and communicate various events,as well as one of the important media for people to obtain news and information.The major domestic epidemic that broke out in December 2019 has aroused widespread attention and discussion on social media.How to accurately analyze the changes in the emotional tendencies of netizens in the event and the concerns of netizens under the corresponding emotional state is crucial for a comprehensive analysis of public opinion.In order to explore the evolution of public opinion in the COVID-19,this article carried out specific research through three parts: data collection,sentiment analysis,and topic extraction:Firstly,design a data collection method that combines breadth-first traversal strategy and webdriver.This article selected the three-month data(from January 2020 to March2020)of the four official Microblogs,including People's Daily,Headline News,Xinhua Viewpoint,and CCTV News,as the research data source.Through the analysis of the data source pages,with the help of the webdriver tool,the crawl task was carried out in a breadth-first traversal manner,and 1,153,545 pieces of text data were obtained.Secondly,propose a sentiment analysis model of Multi-layer Long Short-Term Memory(Multi-LSTM).This model is composed of Bi-LSTM(Bidirectional Long ShortTerm Memory)and LSTM(Long Short-Term Memory)through the neural network sequence,which can more accurately perform sentiment analysis on the pre-processed microblog text information.Based on the analysis results,this paper divided the result set into extremely positive sentiment text set and extremely negative sentiment text set,and respectively visualized the evolution of sentiment values to obtain an emotional conflict evolution map.Finally,propose a topic model combining LDA(Latent Dirichlet Allocation)and TFIDF(Term Frequency-Inverse Document Frequency).Through the analysis of the emotional conflict evolution map,the important time period of the topic research is obtained.The topic model combined with LDA and TF-IDF is used in the text of the corresponding time period to mine the topic words.And perform fine-grained visual display of heat maps for each word to obtain the user's attention point information at each stage of the corresponding emotion.The innovation of this article is to analyze the evolution of public opinion by adopting the idea of first sentiment analysis followed by topic mining for epidemic incidents.This method combines sentiment and themes to produce the effect of "complementary advantages",that is,it can obtain the causes of the corresponding emotion and the user's attention under this emotion based on single sentiment analysis.Also,based on this method,the emotional tendency contained in each topic text can be displayed numerically and visually,which highlights the user's emotional tendency in each topic event.