Climate change is increasing the frequency and severity of natural disasters, and social media has become a primary platform for information dissemination and public emotional expression during such events. The spread of negative information online can lead to cyber violence and adverse psychological outcomes. This study focuses on the 8-13 flash flood in Longcaogou Scenic Area, Sichuan Province, China, an event involving avoidable deaths and child casualties that generated significant online discussion. The high internet penetration rate in China, especially among minors, highlights the importance of understanding and managing emotional contagion on platforms like Sina Weibo. This study aims to model emotional contagion on Sina Weibo, analyzing the spread of positive, negative, and neutral emotions and simulating intervention strategies to mitigate the spread of negative emotions and foster a more positive online environment following such disasters. The research is significant because it provides a framework for understanding and managing the psychological impact of disasters amplified by social media, ultimately contributing to improved disaster response and social governance.

Existing research has explored the classification, description, and prediction of public sentiment on social media during and after disasters. Studies have shown that emotions are infectious and fluctuate over time, influenced by disaster developments and information spread. However, research on the patterns of emotional spread, the intensity of different emotions, and the effectiveness of intervention measures remains limited. While some studies use data-driven models like time series analysis and machine learning, others employ mechanism-driven models such as epidemiological models (SIR, SIRS) and agent-based models to simulate emotion spread. Many studies utilize infectious disease models, adapting them to represent the spread of emotions, but often lack real-world data validation and detailed analysis of intervention strategies. This study builds upon this existing work by developing a more comprehensive model that incorporates key parameters derived from real-world data, allowing for more accurate simulations and evaluations of intervention effectiveness in the specific context of China and Sina Weibo.

This study employed a mixed-methods approach. First, data on posts and comments related to the #Pengzhou flash floods# hashtag on Sina Weibo were collected from August 13th to 20th, 2021, using Python. Data cleaning involved removing stop words, typos, and duplicate information. Sentiment analysis was conducted using GooSeeker's platform to categorize posts and comments into positive, neutral, and negative emotions. Second, a novel emotion-based post-susceptible-comment-removed (PSCR) model was developed. This model, inspired by a COVID-19 transmission model, considers the transmission of emotions from posts to comments, accounting for the different emotional valences of posts and comments, and the time elapsed between posts and comments. Key parameters, such as the proportion of comments infected by different types of posts and the time interval between posts and comments, were calculated using bootstrap methods and exponential distribution fitting. Other parameters (transmission rate, removal rate, initial value of comments) were obtained through curve fitting. Third, a secondary attack rate (SAR) indicator was introduced to quantify transmissibility from posts to comments. Finally, three intervention scenarios were simulated: 1) interrupting transmission paths; 2) changing the number of emotional posts; and 3) changing the numbers of positive and negative posts. The effects of these interventions were evaluated by measuring the cumulative numbers of negative, positive, and neutral comments. A sensitivity analysis was performed on the total number of users (N) parameter. Root Mean Square Error (RMSE) was used to assess the goodness of fit of the PSCR model.

The study revealed that the overall sentiment of posts and comments following the flood was predominantly negative, with three distinct peaks in both post and comment activity. Negative emotions persisted longer and were more contagious than positive emotions. The analysis of the PSCR model's parameters showed that the transmissibility of emotions from neutral posts to comments was higher than from negative or positive posts. Intervention simulations yielded the following results: * **Transmission Interruption:** Interrupting transmission from negative posts was more effective in reducing negative comments compared to interrupting positive post transmission. Similarly, interrupting transmission from neutral posts was most effective in reducing neutral comments. * **Changing the Number of Emotional Posts:** Reducing the number of negative posts by half led to a 14.97% decrease in negative comments and a 7.17% decrease in positive comments. Increasing the number of positive posts increased both positive and negative comments, suggesting a complex interplay between positive and negative emotional responses. * **Simultaneous Changes:** Simultaneously reducing negative and increasing positive posts was most effective at improving the ratio of positive to negative comments. This suggests a combined approach to intervention is beneficial. The sensitivity analysis showed that the model's results were not significantly affected by variations in the total number of users (N), indicating the robustness of the model's findings. The PSCR model demonstrated a good fit to the real-world data (RMSE<51).

This study's findings address the research questions by demonstrating the temporal trends of emotions, the mechanisms of emotional spread, and effective intervention strategies. The results underscore the importance of considering both the volume and valence of online posts in shaping public sentiment during and after disasters. The greater contagiousness and persistence of negative emotions highlight the need for proactive intervention strategies to prevent the escalation of negative public opinion. The effectiveness of simultaneous interventions that reduce negative and enhance positive content underscores the importance of a holistic approach to online public opinion management. The model's relatively good fit to the real-world data validates the PSCR model's potential for practical application in simulating and guiding public sentiment during crises. The findings are relevant to policymakers, social media managers, and public health professionals seeking effective strategies for managing online public opinion during disasters.

This study provides a novel PSCR model for understanding and managing emotional contagion on social media following disasters. The model incorporates parameters derived from real-world data, enabling more accurate simulations and evaluations of intervention strategies. Key findings highlight the greater contagiousness of negative emotions and the efficacy of combined interventions to reduce negativity and promote positivity. Future research could refine the model by incorporating individual-level factors and more granular emotion categories. Furthermore, investigating the effectiveness of different types of interventions beyond those simulated here (e.g., counter-speech campaigns, fact-checking initiatives) would further enhance our understanding of managing online emotional responses during crises.

The study's limitations include focusing on a single event, the potential bias in generalizing findings to other events, and reliance on curve-fitting for some parameters. The model assumes equal susceptibility to emotional contagion among users and does not fully account for individual-level factors or the dynamics between first-level comments and subsequent replies. Future studies should address these limitations by using larger and more diverse datasets and incorporating more nuanced factors into the model.