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.
