Multilingual Cyberbullying Across Social media using BERT

This research addresses a prevalent issue on digital platforms, such as social media and messaging applications: cyberbullying, which is defined as using electronic communication to threaten, harass, or denigrate others. The impact of this behavior can be severe, leading to anxiety, depression, and even suicidal thoughts, often exacerbated by the anonymity of online interactions. To combat this, the proposed system introduces a multilingual cyberbullying detection tool for social networking sites. Leveraging deep learning techniques for content classification, optical character recognition (OCR) for text extraction from images, and natural language processing (NLP) for advanced text analysis, the system effectively detects cyberbullying across languages like English, Hindi, and Telugu. The system utilizes a BERTbased uncased model for detecting cyberbullying in English and Hindi, while a BERT-based multilingual cased model achieves a detection accuracy of 96.32% for Telugu. A user-friendly interface supports evaluation and intervention, and a summarizing tool generates concise incident reports. By integrating diverse datasets and ethical data generation methods, this approach fosters a safer online environment through enhanced robustness and accuracy in cyberbullying detection.