The increasing frequency and intensity of extreme weather events globally necessitate understanding media framing and public perception to inform climate change and adaptation policies. The 2021 Zhengzhou flood in China serves as a case study to examine the interplay between official media reports and public responses on Weibo, China's primary social media platform. This study analyzes posts and comments from both before and after the release of an accountability investigation report in January 2022, aiming to understand the dynamics of information dissemination and public sentiment during and after a crisis. Social media plays a crucial role in disaster communication, serving as both an information source for the public and a tool for authorities to disseminate information and assess public sentiment. Analyzing this communication can improve risk communication efficacy and inform disaster management practices. The study's significance lies in its ability to provide insights into the interaction between official media and the public during a significant disaster, filling a gap in understanding the dynamics of risk communication in the context of state-controlled media.

Existing literature highlights the increasing trend of extreme weather events and their devastating consequences. Studies emphasize the role of social media in disaster communication, with the public using it for information seeking, emotional expression, and help-seeking, while authorities utilize it for risk communication and situation awareness. Research also notes the influence of media framing on public opinion, particularly concerning disaster response and accountability. However, the dynamics between official media and public resonance remain understudied, particularly in contexts with state-controlled media. Previous research has examined media bias in disaster reporting, highlighting potential discrepancies between media narratives and public experience. The study addresses this knowledge gap by analyzing the interaction between official media and public responses in a specific context, leveraging a unique opportunity provided by the post-disaster accountability report.

This study employed a mixed-methods approach, combining quantitative and qualitative analysis of data collected from Weibo. Data included posts from 12 official media accounts and their corresponding public comments. The period was divided into two phases: the emergency response period (July 20, 2021 – January 20, 2022) and the crisis learning period (January 21, 2022 – March 31, 2022), using the release date of the accountability investigation report as a dividing line. Data preprocessing involved removing non-Chinese characters, punctuation, and stop words, followed by word tokenization using the Jieba tool. Latent Dirichlet Allocation (LDA) was utilized for topic modeling, determining the optimal number of topics (k=5) by comparing perplexity across different models. The DUTIR codebook was used for sentiment analysis, categorizing emotions into seven major categories (depression, like, dislike, fear, surprise, joy, and anger). The study then analyzed the distribution of topics and emotions in both the official media posts and the public comments during both periods, comparing the differences and shifts in communication patterns.

During the emergency response period, official media posts primarily focused on factual information regarding the hazard (extreme weather, flood), location (affected areas, metro stations), rescue efforts, the metro failure, and impacts (casualties, damage). Public comments, conversely, emphasized help-seeking (particularly from neglected rural areas), moral support for victims and rescuers, donations, and expressions of condolence. The dominant emotion in official media posts was "like," while public comments were dominated by "anger." Following the release of the accountability investigation report, official media posts centered on the investigation process, punishment of officials, attribution of responsibility for the disaster, and lessons learned regarding both response and prevention. Public comments reflected a mix of emotions, including condolence, praise for those who acted well, condemnation of those responsible, and discussion of lessons learned. The dominant emotion in official media shifted to "depression," while "anger" continued to dominate public comments. Analysis of topic and emotion distribution (Figures 3 and 4) revealed a divergence in focus and emotional responses between the official media and the public across both time periods.

The findings highlight the discrepancies between official media narratives and public concerns. While official media prioritized factual reporting of the major events and the aftermath's investigation, public comments often reflected a wider range of concerns, including the needs of marginalized communities and a stronger expression of anger and grief. The shift in official media sentiment from "like" to "depression" after the accountability report suggests a recognition of systemic failures and potential consequences. The sustained expression of anger from the public underscores the continued need for accountability and effective disaster management. The limited coherence between official narratives and public sentiment has implications for effective crisis communication and disaster response. The study suggests that official communication should be more responsive to the diverse experiences and concerns expressed on social media.

This study provides valuable insights into the communication dynamics during and after a major extreme weather disaster. The divergence in topics and emotions between official media and public comments on Weibo highlights the importance of incorporating public voices into disaster management and policymaking. Future research could explore the effectiveness of different communication strategies in addressing public concerns and fostering trust during and after crises. Further investigation into the role of social media in identifying and addressing the needs of marginalized communities during disasters is also warranted.

The study focuses solely on Weibo data, limiting the scope of public opinion representation. The analysis is confined to the specific context of the 2021 Zhengzhou flood in China, limiting the generalizability of the findings to other disasters or cultural contexts. The use of automated sentiment analysis might not capture the nuances of human emotion perfectly. The short-text nature of Weibo posts limits the depth of analysis regarding narrative strategies.