Towards COVID-19 fake news detection using transformer-based models.

Area of Science:

• Natural Language Processing (NLP)
• Machine Learning (ML)
• Public Health

Background:

• The COVID-19 pandemic amplified the spread of health misinformation, posing significant public health risks.
• Detecting fake news, especially mixed-content articles, presents a complex challenge for NLP.
• Accurate identification of COVID-19 misinformation is critical for public health communication.

Discussion:

• This study evaluates machine learning algorithms, including Bidirectional Encoder Representations from Transformers (BERT) and COVID-Twitter-BERT (CT-BERT), for COVID-19 fake news detection.
• Performance of various downstream neural network structures (CNN, BiGRU) with BERT and CT-BERT models was assessed.
• Model parameters were analyzed in both frozen and unfrozen states to optimize detection capabilities.

Key Insights:

• Fine-tuning pre-trained transformer models like CT-BERT shows high efficacy in identifying COVID-19 fake news.
• Integrating BiGRU layers with the CT-BERT model yielded a state-of-the-art F1 score of 98%.
• The combination of CT-BERT and BiGRU significantly outperforms other evaluated models.

Outlook:

• Advanced ML models offer a promising solution for combating the pervasive issue of online health misinformation.
• Further research can explore hybrid models and larger datasets for enhanced fake news detection.
• These findings support the development of tools to mitigate the impact of health disinformation campaigns.