Quantitative Research on Group Polarization of Bullet-screen Video Websites Based on BERTopic and XGBoost models
Year: 2025
Type: conference-paper
Abstract: The high interactivity feature of bullet-screen video websites has exacerbated the formation of group polarization. Effective analysis of the evolution law of opinions on bullet-screen video websites can help us grasp the public opinion disputes of events more clearly, and thereby provide strong support for public opinion management. Firstly, the BERTopic model is introduced to conduct topic modeling and sentiment calculation for the public opinion related to the event. Combined with the characteristics of bullet-screen video websites, a targeted quantification algorithm for sentiment divergence degree is designed. The XGBoost model is used to fit the sentiment score of video comments with the source prestige, and based on this, the sentiment divergence degree of the event and its sub-topics is calculated. Further carry out the research on the phenomenon of group polarization. Taking the relevant discussions about Chang 'e-6 on the Bilibili website as an example for empirical research, the results show that introducing the degree of emotional divergence can more accurately judge the phenomenon of group polarization and effectively monitor the points of public opinion disputes.
Authors: Jianhua Dai, Yuran Wang
Institution: Nanjing University of Science and Technology
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W7084116601