TTLD: transformer-based threatening language detection in urdu social media text
Scientific Reports, Published online: 23 August 2026; doi:10.1038/s41598-026-67779-1 TTLD: transformer-based threatening language detection in urdu social media text
Organizations prioritize social media activity as a real-time indicator of public sentiment for informed decision-making. However, these platforms also host threatening language at scale, posing direct consequences for public safety. Urdu faces challenges in this domain due to its complex morphology, limited NLP resources, and limited attention from automated threat-detection efforts.
The researchers conducted a benchmark of pre-trained language models, both multilingual and monolingual, on the task. Subsequently, they designed TTLD (Transformer-based Threatening Language Detection) by incorporating XLM-T for domain-adapted contextual embeddings, a Bi-LSTM component for capturing sequential dependencies, and multi-head attention to weight informative input segments.
The resulting model demonstrated exceptional performance on an annotated Urdu tweet dataset, achieving an F1-score of 90.42%, surpassing all baseline models considered. This breakthrough indicates a promising approach for enhanced threat detection in Urdu and other low-resource social media contexts. The study received no external funding or support.
Written by urgent.news from Scientific Reports's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
