How Misinformation and Rebuttals in Online Comments Affect People's Intention to Receive COVID-19 Vaccines: The Roles of Psychological Reactance and Misperceptions

Fun with Frustration? TikTok Influencers' Emotional Expression Predicts User Engagement with COVID-19 Vaccination Messages

This study examined what kinds of TikTok video and message features are associated with user engagement in the context of COVID-19 vaccination. Content analysis was applied to study a sample of 223 COVID-19 vaccination-related videos from creators with at least 10,000 followers. The content analysis involved coding themes, video formats, the valence of attitude toward vaccination, and emotional expressions from the influencers. A majority of videos showcased personal vaccination experiences, followed by fictitious dramas and instructional information. More fictitious dramas expressed unclear attitudes, neither explicitly supporting nor opposing the COVID-19 vaccine, compared to personal vaccination stories and instructional videos. Some imaginative and dramatic scenes, such as zombie transformation or dramatic spasms after taking the vaccines, were widely imitated across influencers, perhaps humorously, and raised concerns about diminishing positive images of vaccine uptake. Videos with simultaneous expression of humor and frustration significantly predicted engagement when the video content opposed or was uncertain about taking the vaccine, implying the effectiveness of mixed emotional attributes within a message. This study provides insight into how social context and message choices by creators interact to influence audience engagement.

Predicting social media rumours in the context of public health emergencies

The spread of rumours on social media in the context of public health emergencies often distorts perceptions of public events and obstructs crisis management. Microblog entries about 28 rumour cases are collected on Sina Weibo during the COVID-19 outbreak. The Modality–Agency–Interactivity–Navigability model is used to identify the key factors of rumour prediction. To investigate the relationship among information modality, information content, information source and rumour identification, the binary logistic regression model is established based on the features of users and microblog entries. In addition, we propose a multi-feature rumour prediction model based on the Bidirectional Encoder Representations from Transformers (BERT) and Extreme Gradient Boosting (XGBoost) models. The proposed rumour prediction model has the best performance compared with other models. The feature importance is then calculated by the SHapley Additive exPlanations (SHAP), which can also explain the XGBoost results. It is shown that the likelihood that microblog entries are rumours decreases as the values of variables such as user influence and the positive sentiment of comments rise. Microblog entries posted on Thursdays or at noon are more probably to be rumours than those posted at other time. The proposed model can assist emergency management departments in establishing a feasible rumour prediction mechanism to guide public opinion against rumours.