An Ensemble Transformers Approach to Detect Bullying in Campus Environments

One of the vices that is becoming increasingly common on school campuses around the world is bullying. Bullying can manifest itself in many forms, such as physical, verbal or social, and can have serious psychological effects on victims. One way to identify potential victims is to analyze their language use, specifically their frequent use of words or phrases associated with anger or sadness. These emotional patterns can serve as indicators of a student’s risk of becoming a bully or victim of bullying. By identifying these patterns early on, schools can prevent potential incidents of bullying. This paper proposes an architecture of ensemble transformers to predict whether a respondent is being bullied or not by transforming text data into basic emotions. Experimental results show that the structure proposed in this paper has certain advantages over existing algorithms in indicators such as F1-score, indicating that it is expected to detect early signs of potential bullying incidents in texts through this method.