New AI framework detects viral pandemic misinformation in real-time
Researchers reveal BERTopic-VP, an AI system flagging harmful health myths 10x faster than traditional methods.
A team of researchers from Mkululi Sikosana, Sean Maudsley-Barton, and Oluwaseun Ajao has introduced BERTopic-VP, a novel AI framework designed to combat health misinformation during pandemics. Published as a preprint on arXiv (arXiv:2608.15691), this tool integrates contextual embedding-based clustering using BERTopic with a post hoc Virality Prioritisation (VP) layer to identify and rank harmful narratives in real time. By combining semantic structure analysis with a logistic propensity-to-spread score, BERTopic-VP can surface low-volume yet high-risk misinformation clusters for human review, enabling early warning systems for public health officials.
The framework was validated on three benchmark datasets—COVID-19_FNIR, Monkeypox, and Constraint—achieving strong classification performance with F1 scores up to 0.950 and ROC-AUC up to 0.989. Unlike traditional topic modeling approaches that treat engagement as a secondary metric, BERTopic-VP prioritizes topics based on both semantic coherence and diffusion potential, even in datasets lacking engagement metadata. The two-stage hybrid misinformation detection module further enhances accuracy by fusing a supervised content-based classifier with verification signals from public health knowledge bases.
- BERTopic-VP combines BERTopic for contextual clustering with a virality prioritization layer to detect misinformation faster.
- Achieves F1 scores up to 0.950 and ROC-AUC up to 0.989 on COVID-19 and Monkeypox datasets.
- Identifies high-impact misinformation clusters using a logistic propensity-to-spread score for early intervention.
Why It Matters
Enables public health agencies to detect and counter harmful health misinformation 10x faster, improving crisis response.