Fig 1.
Full vaccination coverage (1st & 2nd doses) of G7 countries based on data provided by Mathieu et al. [3].
Fig 2.
Changes in first-dose vaccination coverage [3] and number of vaccine-related tweets in Japan.
Fig 3.
Outline of our tweet-selection process.
Fig 4.
An outline of our annotation, learning, and classification process.
Table 1.
Annotation criteria and example tweets.
Example tweets are translated from Japanese.
Fig 5.
Overview of our vaccine-stance classifier.
Table 2.
Vaccination stance dataset.
Table 3.
Comparison of performance of vaccination stance classifiers.
Fig 6.
Changes in number of users with each stance.
Fig 7.
Evolution of polarization of reaction graphs, RWC, and number of users with each stance.
Fig 8.
Changes in user stance distribution.
Table 4.
Transition matrix between users’ initial stances and final stances.
Table 5.
Attributes of referred user accounts.
Fig 9.
Users who were most referred to by neutral-to-pro users (top), remaining-neutral users (middle), and neutral-to-anti users (bottom).
Fig 10.
Users, who passed the chi-squared test of independence, referred to by neutral-to-pro users (top) and neutral-to-anti users (bottom).
Table 6.
Categories of shared external sites.
Fig 11.
External sites which most shared by neutral-to-pro users (top), remaining-neutral users (middle), and neutral-to-anti users (bottom).
Fig 12.
External sites which passed the chi-squared test, shared by neutral-to-pro users (top) and neutral-to-anti users (bottom).
Fig 13.
Changes in keywords in titles of external sites referred to by neutral-to-pro users (top) and neutral-to-anti users (bottom).