The widespread use of the Internet and social media has accelerated the dissemination of information. However, their openness, anonymity, and lack of effective regulation have also contributed to a disorderly online environment and frequent incidents of cyber-violence. This study focuses on comment texts related to cyber-violence incidents and aims to identify and classify three types of discussants: facilitators, mediators, and bystanders.
The study constructs a feature matrix by integrating inherent basic characteristics, objective thematic characteristics, and subjective emotional characteristics. It then employs a range of classification techniques—including traditional machine learning, deep learning, ensemble methods, and hybrid models—to perform group classification of discussants.
The best-performing model achieves an accuracy exceeding 70%, demonstrating the effectiveness of the proposed framework. The study also reveals distinct differences among discussant groups and their respective roles in the propagation of cyber-violence.
This study contributes to the understanding and classification of online discussant groups, enhances the interpretability of classification models, facilitates relationship analysis, uncovers the rules governing public opinion spread, and provides insights for preventing and managing cyber-violence.
