Research On Dynamic Analysis And Predictive Feedback Method Of Online Opinion Based On The Fusion Of Theme And Emotion

In recent years,with the implementation of the "Internet + government affairs" model,the Internet has gradually developed into an important platform for the interaction between the public and the government.The public expresses political demands through the Internet.The government has paid more attention to political interaction in cyberspace to improve the responsiveness of public appeals.Online comment is an important way to express their appeals,including rich emotional information of the public.Emotion is an important driving factor for the development and evolution of public opinion,and its evolutionary mechanism will largely determine the direction and path of public opinion events.In addition,because the Internet has the characteristics of fast spreading speed and wide range of audiences,if information such as event heat and event development tendency is not fed back in time,the best response time will be lost,resulting in the fermentation of public opinion and even affecting the credibility of government departments.Therefore,how to take effective response measures and promptly intervene in management and control is a key issue that needs to be resolved urgently in the government’s response to public opinion crisis.This paper proposes a research method for dynamic analysis and predictive feedback of online public opinion based on the fusion of topics and emotions,and forms the framework of "evaluation-prediction-control" of public opinion.Using the microblog comment of the "Quanzhou Xinjia Hotel Collapse" incident as the data source,the Word2 Vec model was first used to quickly and accurately extract rich semantic features,and the public opinion topics are extracted by TF-IDF + kmeans method to analyze the evolution of public opinion concerns before and after government intervention;The pre-trained language model BERT is used as the word embedding layer,combined with the deep learning model Bi-LSTM to identify the emotional polarity of comments,and analyze the impact of the government response on the topic of public concern and the change of the emotional state of netizens under the interaction between the topic and the emotion;Furthermore,based on the IPA analysis theory,construct a comprehensive analysis matrix of "attention-satisfaction" for each topic of public opinion events,find out the advantages and disadvantages of the government’s crisis public opinion response,grasp the public’s emotional situation,and provide a reference for the government to take effective public opinion response measures;The Richards model is used to draw the change curve of public opinion event scale and emotional polarity intensity in order to predict the development trend of public opinion.On this basis,a feedback model is constructed to provide early warning information and help the government to control public opinion in time.The verification results of the public opinion event of "Quanzhou Xinjia Hotel Collapse" show that the dynamic analysis and prediction feedback method of public opinion proposed in this paper is reasonable and effective,with good explanatory nature,and provides a reference for the government to respond to the public opinion of the Internet crisis.Based on theme and sentiment analysis,an IPA analysis model is constructed to identify the effectiveness of government response measures,combined with the Richards model construction,the principle of control feedback is applied to help the government obtain public opinion warning information and realize the timeliness of public opinion control.Starting from the timeliness and effectiveness of public opinion control,we will help the government improve its ability to respond to crisis public opinion.