Detection of Cyberbullying on Facebook Twitter (X) Using Bi Directional Long Short‑Term Memory and ExtremeGradient Boost Algorithms

Abstract

The social networking sites have transformed digital communication but have simultaneously enabled the escalation of harmful online behaviors, particularly cyberbullying. This recurring formofdigital aggression can lead to serious emotional and psychological harm, including anxiety, depression, and in severe cases, self‑inflicted injury or suicidal behavior. The timely identification and prevention of cyberbullying have become an essential focus of current research. Although numerous machine learning techniques have been applied to detect abusive content, many continue to face challenges such as inefficient kernel tuning, extended training durations, and re duced predictive accuracy. To address these limitations, this study presents a hybrid deep learning architecture that integrates a Bidirectional Long Short‑Term Memory (BiLSTM) network with the Extreme Gradient Boosting (XGBoost) algorithm to improve contextual awareness and classification accuracy. The proposed framework was trained and evaluated on datasets collected from Facebook and X (formerly Twitter), capturing diverse linguistic and behavioral characteristics of user interactions. Experimental results indicate that the BiLSTM–XGBoost hybrid model outperforms conventional classifiers by effectively managing context representation, adaptive learning, and class imbalance. The model achieved 97% accuracy, 95% precision, 92% recall, and an F1‑score of 96%, confirm ing its robustness and efficiency for cyberbullying detection in dynamic social media environments. The study helps educational institutions, online platforms and legal frameworks provide insights into how to better identify cyberbullying in real‑world scenarios. The study’s high recall ensures that cyberbullies are easily identified and it enhances the understanding of how combining multiple models can lead to better performance in cyberbullying detection.