The COVID-19 pandemic prompted numerous countries to implement containment measures, including lockdowns, significantly impacting residents' mental health and well-being. Public reactions to these policies varied, influencing their effectiveness. China's dynamic clearance strategy, characterized by regional lockdowns, initially garnered public support, as seen with the Wuhan lockdown's success in reducing COVID-19 cases and boosting public confidence. However, the Shanghai lockdown marked a turning point, revealing growing public dissatisfaction and highlighting the phenomenon of 'pandemic fatigue'. This study addresses the limited research exploring the spatiotemporal dynamics of this transition and the interplay between public health policies and public sentiment. The study focuses on Wuhan and Shanghai, comparing how public sentiment shifted in these cities in response to pandemic development and policy measures, analyzing the focus of public attention, and exploring the reasons behind this shift. Social media data, particularly Weibo, a major Chinese platform, offers valuable insights into public reactions to containment policies and public concerns. The study draws on life cycle and crisis communication theories, which suggest that public sentiment and topical focus evolve alongside events, and considers the impact of pandemic fatigue, specifically the loss of perceived control over one's environment contributing to discontent. Existing studies, while utilizing social media data, often lack the detailed spatiotemporal analysis and the consideration of individual-level demographic and socio-economic attributes needed to understand the varied outcomes of identical policies across different times and locations. Therefore, this study aims to fill this gap by tracking the interaction between health policies and public sentiment throughout the policy cycle, considering spatial heterogeneity across various groups.

Several studies have explored the relationship between public sentiment and government policies during the COVID-19 pandemic, primarily using social media data. Many studies have focused on the sentiment expressed towards specific events or policies within a specific timeframe or geographical location. However, there’s a lack of research focusing on the co-evolution of containment policies and public sentiment across the entire policy cycle in China, moving from initial public support to widespread negative reactions. Some works have analyzed geographical differences in sentiment using geotagged data, but lack a granular analysis of individual-level socio-economic moderating effects. The researchers also reference relevant theories, including life cycle and crisis communication theories, and the concept of pandemic fatigue as a factor influencing public dissatisfaction with containment measures. This literature review highlights the need for a more comprehensive spatiotemporal analysis of public sentiment towards containment policies, considering both macro-level policy changes and micro-level individual characteristics.

This study collected geolocated Weibo data from Wuhan and Shanghai during their respective lockdown periods, utilizing targeted keywords to select relevant tweets reflecting residents' experiences and emotions. The dataset comprised over 38,000 tweets from Wuhan and over 62,000 from Shanghai. The researchers employed a lexicon-based approach for sentiment analysis, preprocessing the data through word segmentation, stop-word removal, and character conversion. A sentiment ontology dictionary was used to assign sentiment scores to each word, generating daily average sentiment scores for each city. Multiple linear regression (MLR) was employed to examine the impact of various policies (containment, economic, healthcare) on public sentiment, considering factors like cured cases, school closures, workplace closures, public event cancellations, public transport closures, stay-at-home orders, income support, mask mandates, and care for older people. Regression discontinuity design (RDD) was used to identify potential causal relationships for sentiment fluctuations, leveraging time as a breakpoint analysis driver to mitigate confounding factors and estimate policy impact coefficients. Latent Dirichlet Allocation (LDA), an unsupervised learning algorithm, was applied for topic modeling, extracting underlying themes from the Weibo text data to understand prevalent public concerns during different periods. Multiscale geographically weighted regression (MGWR) was utilized to analyze spatial variations in socio-economic factors influencing public sentiment, considering factors like population size, age distribution, number of hospitals and testing sites, GDP, housing price, and building age at the street level. This allowed the researchers to understand the spatial heterogeneity of sentiment and the varying influences of socio-economic predictors across different locations.

Sentiment analysis of Weibo data revealed that while positive sentiment predominated in both Wuhan and Shanghai, significant temporal variations occurred. In Wuhan, positive sentiment peaked around the Lantern Festival, China's National Day of Mourning, and the end of the lockdown. Negative sentiment spiked due to concerns about social justice issues. In Shanghai, positive sentiment peaked on March 17, 2022, while negative sentiment increased after the discussion of university closure policies. Multiple linear regression analysis in Wuhan showed that increased cured cases, containment and closure policies, economic support, and healthcare strategies positively impacted public sentiment. In Shanghai, rising COVID-19 cases correlated with increased negative sentiment, while containment policies, economic policies, and mask mandates had negative effects. Regression discontinuity design (RDD) analysis identified significant shifts in sentiment around key dates in both cities. In Wuhan, sentiment initially improved after lockdown but decreased later. In Shanghai, sentiment declined after the lockdown and the announcement of stringent zero-COVID policies. Topic modeling revealed different focuses of public concern. In Wuhan, initial focus was on infection prevention and family protection, shifting to government policies, life returning to normal, and gratitude for medical staff. In Shanghai, concern was about supplies, reopening, mental health, and containment policies. Multiscale geographically weighted regression (MGWR) analysis showed spatial heterogeneity in negative sentiment, with higher density areas exhibiting higher negativity. Economic variables, housing prices, and building age showed a significant influence on sentiment, revealing a complex interplay of socio-economic factors. Age group also influenced sentiment differently in Wuhan and Shanghai.

The findings highlight a clear shift in public sentiment towards COVID-19 containment policies, moving from initial support to growing dissatisfaction. This shift is influenced by factors such as pandemic fatigue, socio-economic conditions, and the specific policy measures implemented. The different responses in Wuhan and Shanghai reflect the varying contexts of the outbreaks and the evolution of public health strategies. The initial success in Wuhan contrasted with the later negative response in Shanghai, demonstrating the importance of considering the timing and context of policy implementation. The study's comprehensive approach, integrating multiple analytical techniques, provides a richer understanding of the complex interplay between policies, public sentiment, and socio-economic factors. The findings underscore the need for a more nuanced approach to public health policy, considering not only the epidemiological aspects but also the psychosocial and socio-economic impacts on various population segments.

This study provides valuable insights into the dynamic relationship between COVID-19 containment policies and public sentiment in Wuhan and Shanghai. The shift from initial support to growing dissatisfaction emphasizes the need for adaptable public health strategies considering pandemic fatigue and socio-economic factors. The findings suggest incorporating social media monitoring into policymaking and communication, tailoring strategies to specific regions and groups, and developing flexible communication addressing changing public concerns. Future research should explore the long-term effects of pandemic policies and include data from underrepresented groups to ensure a more complete understanding of public sentiment.

The study's reliance on Weibo data might not fully capture the sentiments of less active online demographics, such as older adults or those with limited internet access. Social media data can also be susceptible to biases and does not entirely reflect the full spectrum of public opinion. Future research should incorporate a broader range of data sources and longitudinal studies to mitigate these limitations.