Aspect-based classification of vaccine misinformation: a spatiotemporal analysis using Twitter chatter.

Area of Science:

• Public Health
• Computational Social Science
• Epidemiology

Background:

• Misinformation regarding COVID-19 vaccination on social media poses a significant threat to public safety and hinders global recovery efforts.
• False narratives surrounding vaccines discourage uptake, thereby slowing the return to normalcy and impacting societal well-being.
• Analyzing social media content to detect and understand vaccine misinformation is crucial for developing effective countermeasures and supporting informed decision-making.

Purpose of the Study:

• To develop and validate a framework for analyzing the spatiotemporal progression of vaccine misinformation aspects on social media.
• To identify and quantify common themes within vaccine misinformation to aid stakeholders in decision-making.
• To provide current insights into the spread of misinformation related to various COVID-19 vaccines.

Main Methods:

• Annotation of approximately 3800 tweets into four expert-verified aspects of vaccine misinformation.
• Development of an Aspect-based Misinformation Analysis Framework utilizing the Light Gradient Boosting Machine (LightGBM) model.
• Spatiotemporal statistical analysis of misinformation progression and calculation of Pearson correlation coefficients between misinformation counts and vaccination rates in 43 countries.

Main Results:

• The LightGBM model achieved high classification accuracy, with Area Under the ROC Curve (AUC) of 90.3% for validation and 89.6% for testing.
• Specific aspect accuracies included 87.4% for "Vaccine Constituent," 92.7% for "Adverse Effects," 80.1% for "Agenda," and 82.5% for "Efficacy and Clinical Trials."
• Correlation analysis indicated that 37% of the studied countries experienced reduced vaccine administration due to Twitter misinformation between December 2020 and July 2021.

Conclusions:

• Twitter serves as a valuable platform for understanding the dynamics of vaccine misinformation.
• Machine learning models like LightGBM are effective for multi-class classification of misinformation, demonstrating reliability even with limited social media data.
• The findings underscore the tangible negative impact of online misinformation on public health initiatives, specifically vaccine deployment.