Research On Prediction Of Public Opinion Reversal For Unbalanced Subsets Under The Background Of Big Data

In the era of social media,online public opinion reversal emerges one after another.Accurately and efficiently predicting whether public opinion events will reverse is of great practical significance for preventing public opinion from deteriorating further.Based on the relevant theories of journalism,communication and management,this thesis uses literature research method,case analysis method,investigation method and modeling method to study the prediction of online public opinion reversal.In this thesis,the features of online public opinion reversal events are constructed,the importance of the relevant features is analyzed and the recognition model of online public opinion reversal is constructed.Then,the relevant guidance strategies are proposed from the perspective of public opinion governance.Through research,35 features are proposed from the aspects of netizens,media and government,and the feature correlation coefficient matrix is used to select the important features with the greatest correlation with the reversal of public opinion events for detailed analysis.Furthermore,the feature optimization is carried out based on the evolution process before the reversal point;keywords extraction,sentiment analysis and other techniques are used to analyze the optimized features.In this thesis,the improved KE-SMOTE algorithm has been put forward to balance the event set with great difference in positive and negative sample distribution to ensure the optimal distribution of sample.Then,based on the balanced event set,the integrated classification model of neural network has been built.Finally,the countermeasures are proposed from the perspective of features and prediction of online public opinion reversal events.