Leveraging Transfer Learning to Analyze Opinions, Attitudes, and Behavioral Intentions Toward COVID-19 Vaccines: Social Media Content and Temporal Analysis

"...it is urgent to efficiently collect information on public perceptions to tailor education materials for public and clinical guidance, which will enable primary care physicians to promote COVID-19 vaccines."

Over the past decades, researchers have used social media analytics tools to monitor public sentiment and communication patterns in health crises, such as Ebola and Zika outbreaks. In addition, previous studies have explored knowledge in the context of vaccines using machine learning and deep learning methods. However, several questions related to COVID-19 vaccines remain unexplored: What is the prevalence of user opinions on a social media platform? How many tweets express positive/negative attitudes and behavioural intentions to take vaccines? Which topics are mostly associated with these contents? To answer these questions, this study uses machine learning models and transfer learning models to examine Twitter content expressing user opinions, attitudes, and behavioural intentions toward COVID-19 vaccines. The goal is to support the rollout of COVID-19 vaccines by extracting social media data that can help tailor promotion programmes to fit different populations.

The researchers collected 2,678,372 tweets related to COVID-19 vaccines from 841,978 unique users posted from November 1 2020 to January 31 2021 and annotated 5,000 tweets as the gold standard. They developed machine learning and transfer learning models to classify tweets for three tasks: (1) opinions (yes, no); (2) attitudes (positive, negative, neutral); and (3) behavioural intentions (positive, negative, unknown). The above tasks all focused on COVID-19 vaccines. They then applied the models to predict unlabeled tweets and performed a temporal analysis to capture trends in the unlabeled tweets. In addition, they performed a topic analysis using word clouds and a latent Dirichlet allocation (LDA) model to further understand the content of tweets in the following categories: positive attitudes, negative attitudes, positive behavioural intentions, and negative behavioural intentions. The researchers then identified 10 main topics and relevant terms for each category.

The research revealed that the prevalence of tweets expressing opinions did not change significantly over time. For tweets containing attitudes toward the COVID-19 vaccines, the rate of negative attitudes was 0.754 (95% confidence interval (CI) 0.707-0.795), while the rate of positive attitudes was only 0.246 (95% CI 0.204-0.293). There as a significant change in users' attitudes toward vaccines over time. Among tweets related to behavioural intentions, the rate of tweets indicating that users will not get vaccinated was 0.342 (95% CI 0.229-0.461), whereas the rate of tweets indicating that users will get vaccinated was 0.652 (95% CI 0.539-0.771). There was a substantial increase in the prevalence of tweets expressing positive behavioural intention starting from mid-December 2020. A number of global events happening around that time could explain this increase. For example, a large number of healthcare workers and influential figures received COVID-19 vaccines to increase public confidence. Indeed, social influence has been shown to positively affect the acceptance rate. At the same time, this increase in positive behavioural intentions could generate a positive social influence, which could lead to a higher vaccine acceptance rate.

Key terms identified in the topic modeling in this study could provide guidance to design or optimise vaccine promotion interventions (e.g., education materials). The analysis herein reveals that COVID-19 vaccine promotion strategies need to resolve concerns about side effects and long-term safety issues, virus mutation, and the difference between COVID-19 and the flu. Moreover, promotion strategies should highlight the chance to return to normal life and stay healthy after being vaccinated for COVID-19.

In terms of methodologies, this study demonstrated that transfer learning models outperformed traditional machine learning models in general. In addition, the LDA technique was found to be useful to extract topics from identified tweets.

In conclusion, this paper has provided "a method to automatically analyze the public understanding of COVID-19 vaccines from real-time data in social media, which can be used to tailor educational programs and other interventions to effectively promote the public acceptance of COVID-19 vaccines."