Abstract

The rapid growth of social media platforms has made the automatic detection of online harassment a pressing requirement for safe digital communication. Recent advances in deep learning, including Bi-LSTM and CNN based models, have shown strong results in identifying online hate speech, but most prior studies restrict their evaluation to a small number of explicit, attribute-specific categories. In this work, two Transformer-based architectures, RoBERTa and DistilBERT, are fine-tuned and evaluated on a challenging six-class cyberbullying classification dataset comprising the categories Age, Ethnicity, Gender, Religion, Other_Cyberbullying, and Not_Cyberbullying. RoBERTa achieved the best overall performance, with a test accuracy of 87.79% and a weighted F1-score of 0.88. DistilBERT achieved a comparable test accuracy of 87.19% (weighted F1 = 0.87) while using approximately 47% fewer parameters. An ablation study and a scenario-based evaluation further show that the difficulty is concentrated almost entirely in distinguishing generalised harassment from non-harassment content.

References

• K. Reynolds, A. Kontostathis, and L. Edwards, "Using machine learning to detect cyberbullying," in 2011 10th International Conference on Machine Learning and Applications and Workshops, vol. 2, pp. 241–244, IEEE, 2011.
• M. Dadvar, F. M. de Jong, R. Ordelman, and D. Trieschnigg, "Improved cyberbullying detection using gender information," in Proceedings of the Twelfth Dutch-Belgian Information Retrieval Workshop (DIR 2012), pp. 23–25, Universiteit Gent, 2012.
• A. Bhatt, "Cyberbullying Classification Dataset," Kaggle, 2023. [Online]. Available: https://www.kaggle.com/datasets/andrewmvd/cyberbullying-classification. [Accessed: 2024].
• J. Wang, K. Fu, and C. T. Lu, "SOSNet: A graph convolutional network approach to fine-grained cyberbullying detection," in 2020 IEEE International Conference on Big Data, pp. 1699–1708, IEEE, 2020.
• T. H. Aldhyani, M. H. Al-Adhaileh, and S. N. Alsubari, "Cyberbullying identification system based on deep learning algorithms," Electronics, vol. 11, no. 20, p. 3273, 2022.
• A. Orelaja and O. Akinola, "Attribute-specific cyberbullying detection using artificial intelligence," Journal of Electronic and Information Systems, vol. 6, no. 1, pp. 10–21, 2024.
• J. Patil, V. Patil, K. Prajapati, D. Patel, S. Trivedi, and R. Patel, "Enhanced depression detection on social media using advanced machine learning and linguistic analysis techniques," in International Conference on Intelligent Computing and Communication, pp. 263–275, Singapore: Springer Nature Singapore, 2024.
• H. Kacha, D. Bhikadiya, K. Sharma, and J. Patil, "Comparative analysis of visual motion and multimodal strategies in depression recognition," International Journal of Computer Applications, vol. 187, no. 58, pp. 58–64, 2025.
• J. Patil, K. Prajapati, D. Patel, R. Chauhan, and M. Patel, "A review of transforming AI for depression detection: Transformer model dominance, multimodal approaches, and future pathways," in International Conference on Computing and Machine Learning, pp. 87–106, Singapore: Springer Nature Singapore, 2025.
• J. Patil and J. Sheth, "Deep learning and machine learning approaches for the classification of personality traits," in Rising Threats in Expert Applications and Solutions: Proceedings of FICR-TEAS 2022, pp. 139–146, Singapore: Springer Nature Singapore, 2022.
• J. Patil and J. Sheth, "Comparative study of data sources, features, and approaches for automatic personality classification from text," International Journal of Computer Applications (IJCA), vol. 174, no. 10, 2021.
• J. Patil and J. Sheth, "Data preparation and quality challenges for the personality recognition in Indian languages using machine learning and deep learning approaches," Journal of IoT in Social, Mobile, Analytics, and Cloud, vol. 4, no. 1, pp. 33–40, 2022.
• M. J. A. Patil and M. P. B. Godhwani, "Review of name entity recognition in Marathi language," IJSART, vol. 2, no. 6, 2016.
• S. Verma, J. A. Patil, and I. Tamhankar, "Integrating natural language processing with CGANs to generate customized, realistic traffic scenarios for autonomous vehicle training," in 2024 IEEE International Conference on Smart Power Control and Renewable Energy (ICSPCRE), pp. 1–5, IEEE, 2024.
• D. Pandya, A. Jadeja, M. A. Khan, S. B. Trivedi, M. A. Ramnath, and B. P. Satish, "Significance of sentiment analysis with text-based mining approach," in International Conference on Emerging Trends in Expert Applications & Security, pp. 315–323, Singapore: Springer Nature Singapore, 2024.
• D. D. Pandya, A. Jadeja, S. Degadwala, and D. Vyas, "Ensemble learning based enzyme family classification using n-gram feature," in 2022 6th International Conference on Intelligent Computing and Control Systems (ICICCS), pp. 1386–1392, IEEE, 2022.
• D. D. Pandya, A. Jadeja, S. Degadwala, and D. Vyas, "Diagnostic criteria for depression based on both static and dynamic visual features," in 2023 International Conference on Intelligent Data Communication Technologies and Internet of Things (IDCIoT), pp. 1–1, IEEE, 2023.