OBJECTIVES: The rapid growth of social media has facilitated the spread of mis- and disinformation on vaccines, contributing to more negative public sentiments towards vaccination. This study aims to leverage machine learning-based natural language processing (NLP) algorithms to monitor and analyze vaccine sentiment and hesitancy across three social media platforms.

METHODS: We collected and analyzed social media discussions from 2011-01-01 to 2021-10-31 related to human papillomavirus (HPV) vaccines; measles, mumps, and rubella (MMR) vaccines; and general, unspecified vaccines, from Twitter, Reddit, and YouTube. Our NLP algorithms classify vaccine sentiment as positive, neutral, or negative and further predict vaccine hesitancy aligning with World Health Organizations 3C vaccine hesitancy framework (namely Complacency, Confidence and Convenience). We manually curated a benchmark dataset that includes annotated 30,000 Twitter tweets, 15,000 Reddit posts, and 15,000 YouTube comments. A variety of machine learning algorithms were evaluated. An interactive dashboard was developed to visualize and compare the temporal and geographic trends.

RESULTS: Over 90 million social media discussions were collected. Machine learning-based NLP algorithms have achieved overall accuracy scores ranging from 0.51 to 0.78 on vaccine sentiment prediction and 0.69 to 0.91 on vaccine hesitancy prediction. Temporal trends from the dashboard revealed variations over time in social media activity across vaccine categories regarding vaccine sentiments and hesitancy; for example, sentiments on Twitter for HPV vaccine trended generally more positively than neutrally and negatively over time, whereas MMR vaccine discussions trended more neutrally.

CONCLUSIONS: We developed a social media monitoring tool to track vaccine sentiment and hesitancy using machine learning algorithms. The system could provide real-time temporal and geographical analyses to inform public health actions to improve vaccine update.