COVID-19 Health Beliefs Regarding Mask Wearing and Vaccinations on Twitter: Deep Learning Approach

Area of Science:

• Public Health
• Social Media Analysis
• Machine Learning

Background:

• The COVID-19 pandemic was accompanied by a widespread infodemic, with significant amounts of information and misinformation circulating on social media.
• Prominent organizations like the WHO and CDC, along with influential individuals, provided critical guidance on mitigating COVID-19 transmission.

Purpose of the Study:

• To utilize machine learning and Twitter data from the pandemic era to investigate public health beliefs concerning mask-wearing and vaccines.
• To assess the influence of high-profile cues to action on these health beliefs.

Main Methods:

• Filtered over 646 million COVID-19-related English tweets to create distinct datasets for mask-wearing and vaccines.
• Manually categorized 3500 tweets per dataset based on Health Belief Model (HBM) constructs.
• Trained machine learning models, specifically the XLNet transformer, to classify tweet relevance to HBM constructs, achieving over 81% accuracy.

Main Results:

• Five XLNet transformer models were trained for both mask and vaccine datasets, with each model demonstrating at least 81% classification accuracy.
• Health beliefs related to perceived benefits and barriers were prominent for both mask usage and vaccination.
• The intensity of these health beliefs demonstrated variability, influenced by high-profile cues to action.

Conclusions:

• Health beliefs regarding perceived benefits and barriers, analyzed via Twitter data and machine learning during the COVID-19 pandemic and infodemic, evolved over time.
• These beliefs were responsive to high-profile cues to action disseminated by major organizations and public figures.