Research On The Causes Of Public Opinion Reversal Event Based On Deep Learning From The Perspective Of "Three Element" Type Structure

With the advent of the 5G era,the Internet has become a major venue for exploring topical social events.Social networking sites allow Internet users to freely share their opinions on social hotspots,and the fusion of information from all sides will quickly ignite online public opinion and generate huge radiating influence,which will then lead to difficulties in the supervision,induction and control of public opinion by the relevant departments.Internet users determine the development direction of online public opinion,and online public opinion is a barometer reflecting the emotional attitude of netizens and plays an important role in the healthy and stable development of society.Therefore it is necessary to explore the reasons for the generation of online opinion reversal events.The main work of this thesis is as follows:The difficulty of online opinion analysis lies in how to reduce the time spent on data annotation and achieve accurate prediction of online users’ sentiment in cross-scene opinion events.In order to analyze the mood of opinion events,traditional machine learning and deep learning models must overcome numerous obstacles,such as difficulty in accurately predicting sentiment trends for opinion events associated with different storylines,difficulty in capturing dependency relationships and semantic features,and over-reliance on data annotation.In this thesis,we integrate BERT pre-training and Bilstm-Attention for sentiment analysis of scenario migration,and attempt to migrate the existing annotated opinion event dataset to the "Fito Muxiu Xiu" event dataset for sentiment prediction analysis.The collection and pre-processing of text comment data is the first stage;Building a BERT pre-training model is the next stage to extract word vectors from the representational text;Developing a Bi LSTM-Attention sentiment analysis model in the subsequent stage will enable long-range relationships and semantic features to be captured;and the last step is to use a Softmax classifier to classify sentiment to achieve cross-scene sentiment analysis.Finally,in order to do cross-scene sentiment prediction analysis of public opinion events,a Softmax classifier is utilized for sentiment classification.Based on the "triadic" structure of news communication subjects,the parties involved in the " Fei Tuotuo-Mu Xiuxiu " incident,the voices of mainstream media and the comments of netizens were selected to explore the reasons for the reversal of public opinion from the perspectives of the emotional,content and crowd attributes of public opinion respectively.Firstly,based on the results of the BERT-Bi LSTM-Attention sentiment analysis model,the stages of public opinion development are classified;secondly,the logical relationship between keywords is discussed based on keyword extraction and semantic network analysis,and the influence of opinion leaders’ voices on online users’ sentiment is explored;then,the "double yellow" case,in which all the "triadic" communication subjects play a role,is selected.The "Shuanghuanglian inhibiting the COVID-19" incident,in which all members of the "triad" of communication actors participated,was then contrasted with the "Fei Tuotuo" incident in order to further examine the function of the parties’ responses,the mainstream media,and microbloggers in the formation of public opinion.This was done in order to aid online users in understanding the volatility of public opinion events and to lessen the likelihood of those events being reversed.The goal is to increase social harmony,lessen the rate at which online public opinion events reverse,and aid internet users in understanding how volatile public opinion events may be.