Research On The Dynamic Evolution And Trend Prediction Of Social Media Public Opinion Under Public Emergencie

The online public opinion caused by sudden public events is difficult to predict and can easily trigger negative emotions at the social level,leading to a deterioration of the public opinion environment and adverse effects on management decision-making and response measures.Therefore,exploring the dynamic evolution laws and trends of online public opinion themes in sudden public events,revealing the differences in public discussion content during the public opinion dissemination cycle,can provide scientific basis for guiding public opinion direction and differential prevention and control.In order to explore the evolution law and trend characteristics of online public opinion under public emergencies,this paper crawls the text data about public opinion from microblog,combines the life cycle theory and microblog forwarding data to divide the propagation stage of online public opinion,and uses TF-IDF algorithm and LDA topic model to dig the evolution law and tendency of public opinion topics in the time sequence of events.This article proposes a trend prediction oriented approach to topic evolution from multiple perspectives of public opinion.Based on the external quantitative characteristics of the topic and the internal text,public opinion topics are divided into four levels for evolutionary research,namely dynamic inheritance and evolution between topics,strength evolution of topics,emotional tendency evolution of topics,and evolution between topic words.The topic distribution of public opinion is obtained by using the LDA topic model.By calculating the semantic similarity between topics,the topics with similarity greater than the threshold are connected to form a track to dynamically display the evolution trend of public opinion topics.With the help of the document topic relationship of the topic model,the evolution process of topic intensity in the time series is calculated.By using sentiment analysis technology,the emotional tendencies of each text can be obtained,which can reveal the emotional evolution trends of the general public under different themes.Using a topic dictionary,calculate the word frequency of theme words at different stages,and visualize the evolution of theme words at different cycles using the cooccurrence relationship between theme words.At the same time,this article proposes a topic evolution method based on trend prediction and multiple perspectives of public opinion,taking into account both the external quantitative characteristics and internal semantic characteristics of public opinion topics.For the external quantitative characteristics of public opinion themes,the study combines traditional statistical prediction methods with machine learning methods,using ARIMA as the basic model to fit the linear relationships in the data.Then,the SVR model is used to predict the error sequence between the predicted and actual values of the ARIMA model,fitting the hidden nonlinear laws in the data,and fitting and predicting the trend of changes in the intensity of public opinion themes.Facing internal semantic features,this article obtains theme words through LDA,combines the co-occurrence relationship between theme words,filters out theme words at different stages,and constructs a theme word library.By training the Word2 Vec model,the word vector and semantic distance of each topic word are obtained,and short-term topic word prediction of public opinion at different stages is carried out to analyze the subsequent content of the topic more accurately and intuitively.