EXPLORING TRENDS AND PATTERNS OF CYBERBULLYING THROUGH BIG DATA ANALYSIS

Abstract

Cyberbullying has emerged as a pervasive digital threat affecting 20-40% of youths worldwide, transcending traditional bullying boundaries through its persistent, anonymous, and cross-platform nature. This research addresses critical gaps in cyberbullying detection and pattern recognition through comprehensive big data analysis of 60,233 social media communications across multiple platforms. The study employs three distinct machine learning approaches to develop sophisticated automated detection capabilities: a BERT-based transformer model, a Bidirectional LSTM with attention mechanism, and an Advanced Random Forest classifier with comprehensive feature engineering. The research methodology integrates natural language processing, statistical analysis, and qualitative examination to identify cyberbullying trends, demographic patterns, and linguistic predictors. The BERT model achieved superior performance with 89.86% accuracy and 96.23% AUC, demonstrating exceptional capability in identifying cyberbullying content while maintaining high recall (96.00%) to minimize false negatives. The BiLSTM model provided competitive results (87.57% accuracy, 94.97% AUC) with valuable interpretability features, while the Random Forest approach (82.00% accuracy, 93.19% AUC) revealed critical feature importance insights, identifying average word length (7.68%), sentence count (6.80%), and question count (5.83%) as primary predictors. Feature importance analysis revealed that cyberbullying detection requires sophisticated understanding of structural linguistic patterns beyond explicit content markers, with interrogative patterns and text complexity serving as significant discriminators. The study acknowledges the increased challenges created by the digital transition to COVID-19 and advises strategies at multiple levels based on its evidence-based conclusions. This research enhances theoretical understanding about patterns of digital aggression, and provides practical solutions for a real-time detection system, educational interventions, and policy recommendations. This multidisciplinary approach, which blends perspectives from computer science, psychology, and sociology, improves the capacity to study cyberbullying and offers scalable technological solutions for making online spaces safer.