Improved Classification of Cyber-Bullying Tweets in Social Media Using SVM-MaxEnt Based Dynamic Programming Based Self Organizing Maps

Cyberbullying, an emerging form of bullying facilitated by digital technology, has become a critical issue with the widespread adoption of social media. Manual mechanisms for detecting and mitigating cyberbullying, such as reporting and blocking, have proven ineffective. This research aims to enhance the automatic classification of cyberbullying tweets using advanced machine learning techniques. We utilized a balanced dataset of over 47,000 English tweets labeled by types of cyberbullying, including age, ethnicity, gender, and religion. Our approach integrates feature extraction using SVM-MaxEnt and classification through Dynamic Programming-based Self-Organizing Maps (SOMs). The model successfully identified cyberbullying content with high accuracy and robustness. Results indicate a significant improvement in detection rates, with an accuracy of 92%, precision of 91 %, recall of 90%, and F1-score of 90.5%. This study contributes to the development of more effective cyberbullying detection systems, aiming to protect users from online harassment.