A Study on Cyberbullying Identification Based on Large Model Language Method
Dongjian Song, Zhimei Dong · 2025
In recent years, cyberbullying has led to numerous cases of psychological trauma and fatalities, posing a serious threat to social stability. This paper explores the potential of large language models (LLMs) in detecting and predicting cyberbullying incidents. We developed a cyberbullying detection and prediction system based on an LLM and validated its performance against four classic machine learning models: Support Vector Machine (SVM), Random Forest, Gradient Boosting Decision Tree (GBDT), and XGBoost. Our experimental results demonstrate that the LLM significantly outperforms the comparison models across key metrics. The LLM achieved an accuracy of 93%, with recall rates of 88% for cyberbullying cases and 87% for non- cyberbullying cases. It also attained an Fl score of 84% for cyberbullying cases and 86% for non-cyberbullying cases, alongside precision rates of 89% and 94%, respectively. These findings underscore the effectiveness of LLMs in recognizing and predicting cyberbullying, providing a robust foundation for the development of more accurate and efficient cyberbullying detection systems.