CYBER BULLING DETECTOR FROM MIXED-CODE TEXT

Abstract

Cyberbullying has become a growing concern in Pakistan’s digital landscape, where social media and online platforms have become primary means of communication. The vast linguistic diversity of Pakistan makes it harder for detection systems to identify harmful content throughout the nation. The combination of English with Roman Urdu and Roman Pashto and Roman Balochi and Roman Sindhi in user communications creates challenges for standard content moderation tools that aim to accurately determine offensive language. Online systems designed to detect cyberbullying exist mainly for single-language environments but fail effectively when analyzing code-mixed text because they only use basic keyword filters. The combination of wrong positives with wrong negatives due to this approach diminishes the general success rate of online content moderation systems. Our project presents an NLP-based cyberbullying detection system which specifically addresses the needs of Pakistan's digital environment that uses multiple languages. The system combines rule-based technical methods with scoring procedures to evaluate various regional language text written in Roman script. The system performs real-time detection of harmful content through automated alerts that lead to administrator intervention for proper actions after repeated violation detection. The method provides inclusive accurate and flexible detection of cyberbullying for users across Pakistan’s multilingual online landscape to establish secure digital communication environments.