The COVID-19 pandemic triggered widespread panic buying globally, characterized by shortages of essential goods and disrupted market order. While some studies have explored the psychological and sociological drivers of panic buying, and the role of government intervention in disease prevention and control, there's a significant gap in understanding the effectiveness of specific government interventions in mitigating panic buying during the crucial early stages of an outbreak. Existing research often relies on small sample surveys, structured questionnaires and model simulations, failing to capture the nuanced, immediate public reactions and the full spectrum of information available in online comments. This research aims to fill this gap by analyzing a large dataset of online comments from Hubei province, China, the epicenter of the initial outbreak, to evaluate the impact of different government interventions on panic buying behavior. The study will utilize text analysis techniques to gauge public perception of different interventions and emotional trends to assess their effectiveness.

Existing literature explores panic buying through various lenses, focusing on psychological factors (perceived scarcity, fear), social influences (norms, observational learning), and the role of media (social media, expert opinions, official communication). Studies employ correlation analysis, qualitative analysis, and statistical simulations. Intervention strategies are categorized into psychological, market-based, and network monitoring approaches. Many previous studies focus on model simulations and lack the immediacy of real-time data; they often use structured questionnaires and structural equation models with relatively small sample sizes, overlooking the wealth of information contained in online comments. This paper uses online comment data to address this gap by conducting a more comprehensive analysis of public reactions to government interventions.

This study employs a mixed-methods approach combining data mining, text analysis, and statistical modeling. Data were collected from Weibo, a major Chinese social media platform, using keywords related to panic buying and government interventions in Hubei province. Tens of thousands of comments were scraped and cleaned. The Latent Dirichlet Allocation (LDA) model was employed to identify topics within the comments. The optimal number of topics was determined using a similarity adaptive method based on the average cosine similarity of topics. Sentiment analysis using Python and sentiment dictionaries was conducted to quantify the emotional tone of comments. Semantic network analysis was used to visualize relationships between words and concepts within each category of government intervention (material sufficiency, authority effect, market regulation, and initiative guidance). Finally, multiple regression analysis was conducted to analyze the factors influencing people's emotional responses to government interventions. Variables included the severity of the epidemic, user activity, participation, and influence.

The analysis revealed that news about material sufficiency had the most positive impact on public sentiment, followed by the authority effect, market regulation, and initiative guidance. The study found that government responses were delayed; panic buying began after the lockdown, and information about sufficient supplies was released later. Semantic network analysis showed a strong focus on face masks in both initiative guidance and market regulation categories, highlighting issues of price gouging and counterfeit products. Sentiment analysis revealed that the ‘sufficient materials’ category elicited the highest positive sentiment (47.55% for non-Hubei areas and 56.48% for Hubei), while ‘initiative guidance’ had the lowest (25.26% and 36.55% respectively). The average sentiment scores reflected this pattern. Multiple regression analysis indicated a significant positive correlation between epidemic severity and public perception of government interventions, with epidemic severity alone explaining 60.1% of the variance in public perception; including follow-up comments increased explained variance to 70%. The study also showed that public engagement on the topic peaked within 3-5 hours of the release of government reports.

The findings highlight the importance of timely and effective communication during public health crises. The effectiveness of interventions based on material sufficiency and authority underscores the need for transparent information sharing about resource availability and reliance on credible sources. The delay in government response emphasizes the need for proactive planning and rapid deployment of interventions. The strong public response to market regulation, despite its limitations, suggests a desire for strong government action against price gouging and unethical practices. The positive correlation between epidemic severity and public perception of interventions reinforces the need to tailor communication strategies to the urgency of the situation. The analysis also suggests that combining strategies, such as using authoritative figures to announce the adequacy of supplies, may be more effective than using single approaches.

This study contributes to a deeper understanding of the effectiveness of government interventions during public health crises by utilizing a large dataset of online comments. The findings emphasize the importance of timely information dissemination, the powerful influence of authoritative figures, and the need for robust market regulation. Further research could explore the long-term effects of these interventions, the role of different social media platforms, and the development of more sophisticated models to predict and manage panic buying behaviors. The limitations related to data collection from online comments, the focus on the early stage of the epidemic, and potential internet censorship are noted.

The study is limited by its reliance on online comment data, which may not represent the views of the entire population, particularly those less digitally engaged. The analysis focuses on the early stages of the epidemic and may not be fully generalizable to later stages or other contexts. The potential for internet censorship in China may have affected the data collected. Future research should address these limitations through broader data collection, longitudinal studies, and cross-cultural comparisons.