The COVID-19 pandemic profoundly impacted global health, and social media became a primary source of information and emotional expression. This study focuses on Sina Weibo, China's largest social media platform, to analyze the interplay of emotions expressed by the public, government, and media during the pandemic's initial six months. The research questions addressed are: 1) What were the predominant emotions expressed by each group? 2) What emotional trends were observed over time? 3) What were the temporal relationships between the emotions expressed by these three entities? Understanding these dynamics is crucial for effective crisis communication and for managing the spread of potentially harmful emotions during public health emergencies. The study aims to deepen our understanding of emotional transmission on social media and clarify the relationship between the government, the public, and the media regarding emotional communication during the pandemic, expanding the application of agenda-setting theory to the emotional dimension. The scale of Weibo usage in China during the pandemic—with billions of daily views of pandemic-related information—makes it an ideal platform to study these dynamics. Furthermore, the active role of the Chinese government and traditional media in disseminating information on social media provides a unique context for examining their influence on public sentiment.

Existing literature highlights the significant role of social media in emotional expression and coping during the COVID-19 pandemic. Studies show that exposure to negative pandemic-related information increases the risk of mental health issues, while emotional content significantly impacts individuals and society. Previous research has documented a variety of emotions expressed on social media, including fear, anger, disappointment, and even positive emotions such as hope and faith. While some studies have focused on public emotions, others have examined government responses and media reporting separately. However, there's been a lack of integrated research exploring the interplay of emotions among these three entities on social media during the COVID-19 pandemic. This study aims to fill this gap by examining the types of emotional content, their trends over time, and their dynamic interactions, building upon agenda-setting theory by integrating the dimension of emotional contagion. This suggests that emotions are not merely passively received by the public, but are actively shaped and influenced by government and media messaging.

This study collected data from Sina Weibo, using a web crawler to gather posts containing keywords related to the COVID-19 pandemic. The dataset included 67,689 public posts, 36,740 government posts, and 126,988 media posts, collected from December 30, 2019, to July 1, 2020. Government and media accounts were selected from Weibo's official lists, ensuring representation across different levels of government and media types. The data were cleaned, and stop words were removed using Python's natural language processing tools. Emotional keywords were identified using a lexicon-based approach, leveraging the "emotional vocabulary ontology library" from Dalian University of Technology. A machine-learning model based on the naive Bayes algorithm was used to further analyze emotional tendencies. To examine the dynamic interactions between emotions across the three groups, a vector autoregression (VAR) model was employed, considering several information criteria to determine the optimal lag length. Granger causality tests were conducted to analyze the causal relationships between the emotional expressions of the public, government, and media, and cross-correlation analysis examined their temporal association. Impulse response functions were used to illustrate the sequential influence and causal impact of different subjects' emotions.

Analysis of the emotional content revealed that all three groups—the public, government, and media—predominantly expressed positive emotions on Sina Weibo during the studied period. Specifically, the government exhibited the highest proportion of positive emotions (51.3%), followed by the media (40.6%), and then the public (11.3%). However, the public showed the highest proportion of negative emotions, particularly disgust (18.8%) and sadness (6.8%). A chi-square test confirmed significant differences in emotional categories across the three groups and a correlation among them (Cramer’s V = 0.192, p < 0.001). Analysis of emotional trends showed a general consistency across all three groups: a higher proportion of positive emotions than negative ones, with fluctuations during the initial stages of the pandemic. Negative emotions were prominent in early January, peaking at above 90% for the public, but decreased significantly in late January as the government implemented preventive measures and disseminated information. The VAR model and Granger causality tests revealed significant causal relationships between the emotions of the three groups, showing a pattern of government emotions influencing media emotions, which in turn affected public emotions. The government's emotional expressions had a Granger causal effect on both media and public emotions (p < 0.001 and p = 0.008 respectively). Similarly, media emotions exhibited a Granger causal effect on public emotions (p < 0.001). Cross-correlation analysis further supports these findings, indicating a lag of 13 days between media emotions and public emotions, and a 2-day lead for government emotions over public emotions. The strength of correlation between media and government emotions was especially high, showing that the government's influence on the media was very direct and significant.

The findings support the hypothesis that government emotions significantly influence media and subsequently public emotions. The observed pattern aligns with the unique political structure of China where the government holds considerable influence over media and information dissemination. The study’s extension of agenda-setting theory to incorporate emotional contagion helps illuminate how the government leverages emotional messaging to shape public perception and maintain social stability during crises. The predominantly positive tone from the government and media suggests a deliberate strategy to manage public fear and anxiety, even under the constraint of censorship. While the high proportion of neutral emotions from the public might be partially attributed to censorship, it also indicates a level of social control and adherence to official narratives. The study’s methodologies—the use of VAR models and Granger causality tests—provide a robust and dynamic approach to understanding the intricate causal relationships among different actors in disseminating emotional information.

This study demonstrates a significant influence of government and media emotions on public sentiment during the COVID-19 pandemic on Sina Weibo. The findings extend agenda-setting theory by showing the impact of emotional contagion and highlight the importance of considering emotional dimensions in crisis communication strategies. Future research could explore the interplay between emotions and specific policy support, examining how emotional responses translate into behavioral changes. Further investigation into the role of specific emotional expressions (e.g., fear vs. hope) in shaping public attitudes toward government policies is warranted. The cross-cultural comparison of these dynamics in different political systems would also be valuable.

The study's limitations include the reliance on a lexicon-based method for emotion detection, which may overlook nuances in contextual meaning. The censorship on Weibo might have affected the complete representation of public sentiment, resulting in an underestimation of negative emotions. Furthermore, the focus on emotional aspects neglects the potential interplay with other factors such as information accuracy and dissemination speed. Finally, while Granger causality provides insight into predictive relationships, it does not definitively establish causal links. Future work should address these limitations.