Detecting and monitoring concerns against HPV vaccination on social media using large language models

Health risks due to preventable infections such as human papillomavirus (HPV) are exacerbated by persistent vaccine hesitancy. Due to limited sample sizes and the time needed to roll out, traditional methodologies like surveys and interviews offer restricted insights into quickly evolving vaccine concerns. Social media platforms can serve as fertile ground for monitoring vaccine-related conversations and detecting emerging concerns in a scalable and dynamic manner. Using state-of-the-art large language models, we propose a minimally supervised end-to-end approach to identify concerns against HPV vaccination from social media posts. We detect and characterize the concerns against HPV vaccination pre- and post-2020 to understand the evolution of HPV vaccine discourse. Upon analyzing 653 k HPV-related post-2020 tweets, adverse effects, personal anecdotes, and vaccine mandates emerged as the dominant themes. Compared to pre-2020, there is a shift towards personal anecdotes of vaccine injury with a growing call for parental consent and transparency. The proposed approach provides an end-to-end system, i.e. given a collection of tweets, a list of prevalent concerns is returned, providing critical insights for crafting targeted interventions, debunking messages, and informing public health campaigns.