TL;DR
This study analyzes social media information behavior during the 7.20 Henan heavy rain flood disaster, revealing a distinct life cycle of information engagement and significant variations based on location, content, and user demographics. Understanding these patterns is crucial for effective disaster risk governance and communication strategies in the digital age.
What Problem Does This Paper Address?
Natural disasters pose significant threats to social-ecological systems, and how information spreads and is consumed during these events is critical for disaster response and mitigation. While disaster information behavior is recognized as a key factor, research has largely overlooked the specific dynamics of major natural disasters and their manifestation on social media. This paper aims to fill this gap by investigating the characteristics and differences in disaster information behavior on social media, using the 7.20 Henan heavy rain flood disaster as a case study. The goal is to provide a scientific understanding of how information dissemination impacts disaster prevention and to identify factors influencing this behavior for better integrated disaster risk governance.
Methods
The researchers employed a multi-faceted approach to analyze disaster information behavior. They constructed an "information behavior composite index" by combining data on Weibo posts, forwards, and comments over a three-day rolling window, then standardized this index to a 0-1 scale. To understand the temporal dynamics, they analyzed precipitation data and the information behavior index over time, identifying distinct phases. Spatial analysis mapped the distribution of information behavior across different regions. Content analysis categorized microblog posts into "hazard factors," "disaster loss and exposure," and "disaster response/relief," with further sub-categorization within the latter. Finally, behavioral agents were analyzed by examining user demographics (gender and age) and the sentiment of their comments, using a naive Bayesian algorithm for sentiment analysis. Statistical methods like Granger causality analysis (GCA) and impulse response analysis were used to explore relationships between variables.
Datasets
The study utilized two primary datasets:
- Precipitation Data: Daily precipitation data for Henan Province and Zhengzhou from the China Meteorological Data Service Centre Daily Value Dataset (V3.0). This data was used to track the hazard factor of the disaster.
- Social Media Data: Microblog data (posts, forwards, comments) related to the "7.20 Henan heavy rain flood disaster" from Sina Weibo, collected between July 12 and September 5, 2021. This dataset comprised 25,590 popular comments and 16,922 microblogs.
Key Insights and Ideas
- Disaster Information Has a Life Cycle: Information behavior isn't static; it follows a predictable "formation-development-extinction" pattern, peaking before the peak of normal distribution, suggesting a rapid initial surge and then a decline.
- Location Matters: Information engagement is higher in disaster-stricken areas and economically developed regions, highlighting the influence of proximity and resources on information dissemination.
- Content is Mixed: A significant portion (53%) of social media content was unrelated to the disaster, indicating that while disaster response was a major theme (43.08%), personal interests and unrelated topics also heavily populate the discourse.
- Demographics Drive Behavior: Females are more active and tend to post more positive content (e.g., "SOS & rescue," "Praying for safety"). The 20-29 age group is the most active, while the 50+ group is the least. Sentiment also varies by age, with younger groups showing more positive/neutral comments and older groups leaning towards negative/irrelevant.
- Psychological Distance is Key: The closer people are spatially and socially to a disaster, the more engaged they are, suggesting that personal experience and perceived risk significantly shape information behavior.
Key Results
- Temporal Pattern: Information behavior exhibits a clear life cycle with three phases: latent, outbreak, and receding. The peak of information behavior occurred one day after the peak precipitation in Zhengzhou.
- Spatial Distribution: Information behavior levels were higher in disaster-affected areas and economically developed provinces (e.g., Guangdong, Shandong) compared to less developed regions.
- Content Analysis: 47% of microblog content was related to the disaster. Of the related content, "Disaster response/relief" was the most frequent category (43.08%), followed by "Disaster loss and exposure" (41.54%). Within "Disaster response/relief," "Prewarning & forecasting" (21.19%) and "Donation" (20.21%) were prominent.
- Behavioral Agents: Females constituted 54.19% of information behavior, compared to males at 45.81%. The 20-29 age group showed the highest information behavior level (45%), while the 50+ group had the lowest (1.96%). Positive comments were most prevalent among those under 19, neutral comments among 20-29 year olds, negative comments among 30-39 year olds, and irrelevant comments among 40-49 year olds.
Limitations and Future Work
The authors acknowledge several limitations:
- Distinguishing Disaster vs. Non-Disaster Areas: The study did not explicitly distinguish between information behavior in disaster-affected areas versus non-disaster areas.
- Impact on Natural Disaster Events: The direct impact of information behavior on natural disaster events themselves was not discussed.
Future work could focus on:
- Investigating the mechanisms and pathways by which information dissemination stimulates and regulates the action of natural disaster events on social systems.
- Further exploring the differences in information behavior between disaster and non-disaster areas.
Summary updated