When Cyber Aggression Prediction Meets BERT on Social Media
Increasingly, cyber aggression becomes the prevalent phenomenon that erodes the social media environment. However, due to subjective and expense, the traditional self-reporting questionnaire is hard to be employed in the current cyber area. In this study, we put forward the prediction model for cyber aggression based on the cutting-edge deep learning algorithm. Building on 320 active Weibo users' social media activities, we construct basic, dynamic, and content features. We elaborate cyber aggression on three dimensions: social exclusion, malicious humour, and guilt induction. We then build the prediction model combined with pretrained BERT model. The empirical evidence shows outperformance and supports a stronger prediction with the BERT model than traditional machine learning models without extra pretrained information. This study offers a solid theoretical model for cyber aggression prediction. Furthermore, this study contributes to cyber aggression behaviors' probing and social media platforms' organization.