Enhancing public health response: a framework for topics and sentiment analysis of COVID-19 in the UK using Twitter and the embedded topic model.

Abstract

To protect citizens during the COVID-19 pandemic unprecedented public health restrictions were imposed on everyday life in the UK and around the world. In emergencies like COVID-19, it is crucial for policymakers to be able to gauge the public response and sentiment to such measures in almost real-time and establish best practices for the use of social media for emergency response. In this study, we explored Twitter as a data source for assessing public reaction to the pandemic. We conducted an analysis of sentiment by topic using 25 million UK tweets, collected from 26th May 2020 to 8th March 2021. We combined an innovative combination of sentiment analysis via a recurrent neural network and topic clustering through an embedded topic model. The results demonstrated interpretable per-topic sentiment signals across time and geography in the UK that could be tied to specific public health and policy events during the pandemic. Unique to this investigation is the juxtaposition of derived sentiment trends against behavioral surveys conducted by the UK Office for National Statistics, providing a robust gauge of the public mood concurrent with policy announcements. While much of the existing research focused on specific questions or new techniques, we developed a comprehensive framework for the assessment of public response by policymakers for COVID-19 and generalizable for future emergencies. The emergent methodology not only elucidates the public's stance on COVID-19 policies but also establishes a generalizable framework for public policymakers to monitor and assess the buy-in and acceptance of their policies almost in real-time. Further, the proposed approach is generalizable as a tool for policymakers and could be applied to further subjects of political and public interest.