Classifying COVID-19 Vaccine Narratives

Proceedings of Recent Advances in Natural Language Processing, pages 648–657 Varna, Sep 4–6, 2023 https://doi.org/10.26615/978-954-452-092-2_070 648 Classifying COVID-19 Vaccine Narratives Yue Li, Carolina Scarton, Xingyi Song and Kalina Bontcheva Department of Computer Science, University of Sheffield (UK) {yli381, c.scarton, x.song, k.bontcheva}@sheffield.ac.uk Abstract Vaccine hesitancy is widespread, despite the government’s information campaigns and the efforts of the World Health Organisation (WHO). Categorising the topics within vaccine- related narratives is crucial to understand the concerns expressed in discussions and identify the specific issues that contribute to vaccine hesitancy. This paper addresses the need for monitoring and analysing vaccine narratives online by in- troducing a novel vaccine narrative classifica- tion task, which categorises COVID-19 vaccine claims into one of seven categories. Following a data augmentation approach, we first con- struct a novel dataset for this new classification task, focusing on the minority classes. We also make use of fact-checker annotated data. The paper also presents a neural vaccine narrative classifier that achieves an accuracy of 84% un- der cross-validation. The classifier is publicly available for researchers and journalists. 1 Introduction Vaccination is one of the most effective public health interventions, but it is essential that immu- nisation programs are able to achieve and sustain high vaccine uptake rates. Overcoming vaccine hesitancy, which refers to the delay in the uptake or refusal of vaccines, is a major challenge (Eskola et al., 2015) and the WHO has named it one of the top ten threats to global health in 2019 (Qayum, 2019). Vaccine hesitancy is a complex and context specific phenomenon, varying across time, place and even vaccines (Larson et al., 2014). It could be caused by various factors such as concerns about side effects, costs, and misinformation. Although social media platforms like Twitter, Facebook, and YouTube have taken actions to limit the spread of misinformation, simply identifying and removing misinformation from platforms is not enough, as the concerns of the vaccine-hesitant citizens also need to be monitored and responded to. Consequently, fact-checkers and other pro- fessionals need analytical tools that help them to better monitor misinformation, vaccine hesitancy, vaccine-related debates and their narratives. Topic analysis of narratives about vaccines could be used for this purpose, however, a large man- ual effort is required, due to the lack of a vaccine- related topic classifier. For example, Smith et al. (2020) gather over 14 million vaccine-related posts from Twitter, Instagram, and Facebook to research vaccine-related narratives. The posts are cate- gorised into six topics based on a novel typology designed to capture the ways narratives are framed. However, manual analysis was feasible on only a small sample of 1,200 posts, which, given the small scales, leaves significant gaps in the understanding and tackling of vaccine hesitancy. Guided by these needs, the novel contributions of this paper are in: 1. Proposing a new seven-way classification task and dataset for categorising vaccine re- lated online narratives. The classification task adopts the six categories (see Table 1) defined by Smith et al. (2020). The dataset is built based on …

Content truncated to 500 words