A culturally aware NLP approach for fake-news detection in Nigerian online media using BERT

Abstract

Fake-news dissemination in multilingual Nigerian media poses serious sociopolitical risks, yet many existing detection models do not adequately account for Nigeria's linguistic and cultural diversity. This study presents a culturally aware natural language processing (NLP) model for fake-news detection based on the Bidirectional Encoder Representations from Transformers (BERT) architecture. A curated corpus of 10,000 Nigerian news articles was assembled and stratified across Standard English, Nigerian Pidgin, Yoruba, Hausa, and Igbo. The model combines language-adaptive fine-tuning with regulatory-informed annotation based on the standards of the National Broadcasting Commission (NBC), Nigerian Communications Commission (NCC), and National Information Technology Development Agency (NITDA). Experimental results show that the proposed model achieved 91.3% accuracy, 90.8% precision, and an 89.7% F1-score, outperforming vanilla multilingual BERT and traditional machine-learning baselines. The system also performed strongly on indigenous-language and code-switched content, with an average inference latency below 400 ms. The study provides a scalable and context-dependent decision-support model for real-time misinformation detection in the Nigerian online-media environment.