A comparative analysis for spatio-temporal spreading patterns of emergency news

Emergency events significantly impact economic and social development. Effective communication is vital during emergencies to inform affected populations and coordinate rescue and recovery efforts. While traditional communication networks can fail, social media platforms like Sina Weibo offer valuable alternative channels for information dissemination. Existing studies often focus on specific events, limiting generalizability. This research aims to develop a framework for quantifying general spreading patterns of emergency-related information and comparing them to other news types to provide insights for improved emergency management strategies. The increasing use of social media during emergencies necessitates understanding the unique dynamics of information transmission, requiring novel methods to quantify the patterns and unique evolution of emergency-related information spread across social media platforms. This study addresses this gap by developing a systematic framework to analyze large-scale social media data to quantify and compare the spreading patterns of emergency news with other news categories. The framework will provide insights into the characteristics of emergency news dissemination, ultimately contributing to more effective emergency response and crisis communication.

Several studies have explored the use of social media data during emergencies, focusing on identifying community needs and connecting affected populations with external organizations. Research has also examined disaster recovery using social media data to create aggregate pictures of population response. Additionally, existing work has investigated information transmission in online social networks, analyzing user emotions, behaviors, and response differences during emergencies. However, many studies are limited to specific events, lacking generalizability, and do not address critical questions about the duration of information dissemination cycles, the timeline of public opinion formation and recovery, and the spread of accurate information amidst misinformation. This paper addresses these gaps by providing a comprehensive comparative analysis.

This study collected data from Sina Weibo, China's largest microblogging platform, focusing on 193 popular events classified into five categories: emergency, entertainment, political, social, and tech news. The dataset comprised 81,606,262 forwarding records from 3,514,660 users. The researchers developed a comprehensive set of measures to quantify the communication effect of microblog news: **Temporal Measures:** * **Life Cycle:** The total duration of news propagation. * **Active Period:** Duration of high popularity. * **Fluctuation:** Number of significant change points in daily reposts (using Pettitt's test). * **Inter-person Diffusion Time:** Time between receiving and reposting information. **Network Topology Measures:** * **Propagation Tree:** A directed graph illustrating information spread. * **Depth:** Maximum number of levels in the propagation tree. * **Size:** Number of unique users participating. * **Breadth:** Maximum number of nodes at any level. * **Propagation Modes:** Classified into Ripple, Dandelion, Colony, and Firework modes. **User Engagement Measure:** * **Efficiency:** Ratio of unique users to total reposts. **Information Coverage Measure:** * **Covered Users:** Sum of followers of all reposters. Data processing involved manual classification of news into five categories and the application of the above measures. Propagation mode classification used a logistic regression model based on breadth features. An algorithm based on binary search was used to refine the classification model. The analysis compares the measures across the five news categories.

The study revealed significant differences in the spread patterns across the five news categories: **Temporal Evolution:** * Emergency news had significantly shorter lifecycles (median 15 days) and active periods (median 1.78 days) than other news types. * Emergency news showed fewer post-peak fluctuations (over 62.9% experienced no significant change). * The inter-person diffusion time for emergency and tech news was shorter than for other news types. * Emergency news reposting peaked between 8 am and 4 pm. **Network Propagation Characteristics:** * Emergency news had shallower propagation depths (median 2 levels) and narrower breadth (median 1000) compared to other news types. * Emergency news predominantly followed a ripple propagation mode (79%), spreading primarily within the original poster's network. **User Engagement:** * Emergency news showed high repost efficiency (70-80%), indicating fewer repeated reposts by the same users compared to other news types. **User Influence:** * Unlike other news types, the number of reposts for emergency news was not strongly correlated with the number of followers or fans of the reposting users. **Information Coverage:** * Linear fitting showed that emergency news exhibited high significance in the relationship between reposts and the number of covered users. The model showed that the original publishers of emergency news had a larger average spread of influence than social and tech news. An ensemble learning regression model was developed to predict information coverage using all the features, which demonstrated improved prediction accuracy compared to other individual models.

The findings demonstrate that emergency news dissemination exhibits unique spatio-temporal patterns compared to other news types. The shorter lifecycles, active periods, and limited propagation depth of emergency news highlight the time-sensitive nature of emergency information. The dominance of the ripple propagation mode suggests that initial information sources play a crucial role in spreading emergency news, emphasizing the importance of well-established and trusted channels. The low correlation between user influence and reposts for emergency news implies that strategies focusing on influencer marketing might be less effective compared to other news types. The successful prediction of information coverage using the ensemble learning model underscores the potential of the proposed framework in assessing and improving crisis communication strategies. These findings offer insights into how to improve the speed and effectiveness of delivering crucial information during emergencies.

This study identifies unique spreading patterns of emergency news on Sina Weibo. Emergency news exhibits shorter lifecycles, active periods, and fewer fluctuations compared to other news types, with a primarily ripple propagation mode. User influence has limited impact on dissemination. An ensemble learning model successfully predicts information coverage. Future research should explore causal factors behind these patterns and refine prediction models.

This study is limited to data from Sina Weibo, a Chinese microblogging platform. The findings might not be directly generalizable to other platforms or cultures. The manual classification of news categories could introduce some subjective bias. Further research could explore a broader range of events and platforms to enhance the generalizability of the findings.