An Explainable Transformer-Based Framework for Bangla Health Misinformation Detection on Social Media
Abstract
The widespread circulation of health misinformation on Bangla social media poses a serious public health concern. To address this issue, an explainable transformer-based framework is proposed for Bangla health misinformation detection. A dataset of 5,038 Bangla health-related statements has been developed from various online platforms such as social media, blog and annotated as REAL or FAKE based on verification by medical and food safety specialist. The dataset covers misinformation on major diseases including cancer, heart attack, stroke, dengue, malaria, seasonal flu, and COVID-19. Four transformer modelsBERT, mBERT, XLM-RoBERTa, and BanglaBERThave been fine-tuned and evaluated on this dataset. BanglaBERT has achieved the best performance with <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 6 \%}$</tex> accuracy, precision, recall, and F1-score, outperforming other multilingual and cross-lingual models. The proposed framework has demonstrated its effectiveness and reliability for automated Bangla health misinformation detection on social media platforms.