Tapaopong W.Charoenphon A.Raksasri J.Samanchuen T.Mahidol University2024-09-052024-09-052024-01-015th Technology Innovation Management and Engineering Science International Conference, TIMES-iCON 2024 - Proceedings (2024)https://repository.li.mahidol.ac.th/handle/20.500.14594/100985—The proliferation of cyberbullying on social media platforms has led to devastating physical, emotional, and mental consequences for victims, underscoring the need for effective identification and analysis methods. This study leverages Natural Language Processing (NLP) techniques, specifically Transformer models, to enhance cyberbullying detection in harmful social media messages. By employing transfer learning and fine-tuning strategies, we compared the effectiveness of category prediction using five prominent Transformer models: BERT, RoBERTa, ALBERT, DistilBERT, and ConvBERT. Our results demonstrate that DistilBERT outperforms other models, achieving an accuracy of 94.36%, precision of 93.83%, and recall of 93.91% in the 5-Class categorization of cyberbullying severity. This study highlights the significance of automated cyberbullying detection and underscores the potential of NLP techniques in combating the adverse impacts of cyberbullying. The findings of this research have important implications for the development of effective cyberbullying detection systems, which can help mitigate the harmful effects of online harassment. By leveraging the capabilities of DistilBERT, we can create more accurate and efficient systems for detecting and preventing cyberbullying, ultimately contributing to a safer online environment.EnergyBusiness, Management and AccountingComputer ScienceMedicineDecision SciencesConference PaperSCOPUS10.1109/TIMES-ICON61890.2024.106307192-s2.0-85202600754