Opinion Change Towards COVID-19 Vaccine on Japanese Twitter

Dynamic Community Detection and Opinion Mixture Analysis

Abstract

Since the outbreak of the COVID-19 in 2020, discussions surrounding vaccines have consistently ignited public discourse on social media. Previous studies related to the social networks discussing the vaccine topics have illuminated the existence of distinct clusters. However, these studies have not delved into the dynamic shifts in these communities and their opinions. They also lack a deep understanding of opinion mixtures inside each community. Our objective is to track the evolution of social communities related to vaccine and scrutinize the dynamic opinions in each community. To accomplish this, we gathered a dataset comprising 45 million tweets and 80 million retweets using the Twitter API, employing search criteria that included the keyword "vaccine (in Japanese)" and a time frame spanning from January 2022 to June 2022. To begin with, we constructed a retweet network and then applied the Louvain method which detected 6 primary communities. Subsequently, we examined the evolution of these communities on quarterly basis. We also trained an opinion classifier using supervised learning (80% precision) which help us understand the different opinion mixture and its changes in each community.This comprehensive analysis offers valuable insights into the evolution of communities and their contributions to shifts in collective attitudes towards vaccines.