Identifying Disinformation on the Extended Impacts of COVID-19: Methodological Investigation Using a Fuzzy Ranking...

Area of Science:

• Artificial Intelligence
• Natural Language Processing
• Public Health Informatics

Background:

• The COVID-19 pandemic highlighted the pervasive threat of online misinformation to public trust and understanding of health policies.
• Persistent misinformation regarding long-term COVID-19 effects and reinfection necessitates integration into public health strategies.

Purpose of the Study:

• To develop a robust and generalizable deep learning framework for detecting misinformation on prolonged COVID-19 impacts.
• To integrate pretrained language models (PLMs) with a novel fuzzy rank-based ensemble approach for enhanced detection accuracy.

Main Methods:

• Curated a dataset of 566 genuine and 2361 fake COVID-19 misinformation samples.
• Employed state-of-the-art PLMs (RoBERTa, DeBERTa, XLNet) within an ensemble strategy.
• Integrated a reparameterized Gompertz function for fuzzy rank-based assignment of model prediction confidence.

Main Results:

• XLNet demonstrated superior performance among individual language models due to its permutation language modeling.
• The fuzzy rank-based ensemble method achieved high performance metrics: 93.52% accuracy, 94.65% precision, 96.03% F1-score, and 97.15% AUC.
• Model architecture, training, and optimization significantly influence classification effectiveness, with larger parameter models generally outperforming others.

Conclusions:

• The fusion of ensemble learning with PLMs and the Gompertz function offers a novel, accurate, and reliable prediction approach for misinformation detection.
• High prediction accuracy is achievable using solely textual content, providing insights for optimizing fake news detection systems.
• Findings have broader implications for applying deep learning in public health policy and communication to combat misinformation.