Table 1.
Sample keywords used for collecting COVID-19 and vaccine-related tweets.
Table 2.
Categorization of collected tweets.
Table 3.
Classification of vaccine stance in tweets.
Table 4.
Thematic categorization of anti-vaccine tweets.
Table 5.
Criteria for stance consistency classification in quote tweets.
Fig 1.
Temporal distribution of vaccine stance tweets before and after policy change.
Each color represents a distinct stance category: anti-vaccine, neutral, pro-vaccine, and no mention of vaccine stance. The dashed line denotes the policy change date (Nov 23). The y-axis shows the percentage of each stance category, highlighting the relative shift in composition following the policy update.
Fig 2.
Temporal trends in vaccine stance tweets by tweet type.
These three subplots illustrate vaccine stance trends across original tweets (TW), retweets (RT), and quote tweets (QT). The dashed line represents the policy change date. The y-axis shows the percentage of each stance category within each tweet type.
Fig 3.
Percentage distribution of vaccine stances across tweet types before and after policy change.
Solid bars represent pre-policy data, while hatched bars indicate post-policy data. Percentages are shown above each bar, facilitating direct comparison of compositional shifts across tweet types.
Table 6.
Examples and distribution of non-anti-vaccine tweets before and after policy termination.
Table 7.
Anti-vaccine tweet topics, distribution, and examples before and after policy termination.
Fig 4.
Percentage distribution of anti-vaccine themes.
The most prevalent theme is health concerns, followed by against mandatory vaccination and deeper conspiracy. Less common topics include shedding and Big Pharma. Percentages are shown above each bar, illustrating the relative prevalence of vaccine hesitancy narratives.
Fig 5.
Time series of anti-vaccine tweet topics from Nov 16 to Nov 30.
Each color represents a different theme, and the dashed line marks the policy change date (Nov 23). The y-axis shows the percentage composition of topics, revealing how the relative prominence of health concerns, government conspiracies, and vaccine inefficacy narratives shifted following the policy change.
Fig 6.
Association between policy termination and anti-vaccine tweets.
Odds ratios with 95% confidence intervals are presented for different tweet types. The overall population exhibits a significant increase in anti-vaccine tweets, while quoted tweets show no statistically significant change.
Table 8.
Examples of highly retweeted anti-vaccine tweets during the study period.
Fig 7.
Top 20 most retweeted tweets during the study period, colored by vaccine stance.
Hatched bars indicate tweets from the post-policy period (Nov 24–30). Despite having shorter observation windows for accumulating retweets (0–6 days versus 8–14 days for pre-policy tweets), post-policy tweets comprised 50% of the top 20 most retweeted tweets, suggesting accelerated spread of anti-vaccine content following policy termination.
Fig 8.
Odds ratios for anti-vaccine themes across tweet types before and after policy change.
Each panel corresponds to a distinct anti-vaccine narrative, showing whether policy change was associated with its prevalence. Statistically significant increases and decreases are observed for various tweet types.
Fig 9.
Distribution of vaccine attitudes against standpoint consistency before and after policy change.
The leftmost heatmap represents the overall distribution, while the middle and right heatmaps correspond to the pre-policy and post-policy periods, respectively. The color intensity reflects the frequency of each category, with darker shades indicating higher tweet counts. Notably, anti-vaccine quote tweets show increased alignment with the quoted content after the policy change.