With the increasing use of social media, online self-organized relief has become a crucial aspect of crisis management during public health emergencies, leading to the emergence of online self-organizations. This study employed the BERT model to classify the replies of Weibo users and used K-means clustering to summarize the patterns of self-organized groups and communities. We then combined the findings from pattern discovery and documents from online relief networks to analyze the core components and mechanisms of online self-organizations. Our findings indicate the following: (1) The composition of online self-organized groups follows Pareto's law. (2) Online self-organized communities are mainly composed of sparse and small groups with loose connections, and bot accounts can automatically identify those in need and provide them with helpful information and resources. (3) The core components of the mechanism of online self-organized rescue groups include the initial gathering of groups, the formation of key groups, the generation of collective action, and the establishment of organizational norms. This study suggests that social media can establish an authentication mechanism for online self-organizations, and that authorities should encourage online interactive live streams about public health issues. However, it is important to note that self-organizations are not a panacea for all issues during public health emergencies.
Keywords: emergency response; machine learning; online self-organization; patterns and mechanism; public health emergencies; social media use.