Research on the Prediction of Public Opinion Reversal Based on Machine Learning in New Media Context

Abstract

With the Internet development in the new media era, frequent public opinion reversal events have been a challenge that cyberspace governance should tackle urgently. Predicting the public opinion reversal events in time can help relevant departments grasp the trend of public opinion and formulate effective coping strategies. Based on the typical public opinion events on Weibo from 2022 to 2024, this paper extracts multi-dimensional variables such as communication characteristics, agenda setting degree, and emotional tendency. Also, two text vectorization methods are adopted for calculation: TF-IDF and Doc2Vec, and machine learning methods are introduced for modeling prediction and evaluation. According to the empirical results, Doc2Vec performs better than TF-IDF, and Doc2Vec’s deep learning model performs best, which can improve the recognition accuracy of public opinion reversal events. The research offers technical insights into constructing a prediction model of public opinion reversal, and supports the development of public opinion prediction.