Public perception of earthquake events: evidence from social media – a case study of the 2025 Dingri earthquake

Social media have become the core channel for postearthquake information dissemination and public emotional expression. Focusing on the 2025 Tibet Dingri 6.8-magnitude earthquake, Sina Weibo posts related to the event are obtained, and a ‘topic-sentiment-damage’ three-dimensional analytical framework is constructed: BERTopic is used to extract public discussion topics, the SKEP model is used for topic–emotion coupling, a damage classifier is applied to quantify textual physical damage levels, and a cumulative damage index is calculated within two-hour windows. The findings show pronounced spatial differences in information propagation, with discussion hotspots concentrated in major cities far from the epicenter, whereas voices near the epicenter are marginalized due to weak digital infrastructure. Public discussions focus on disaster events, humanitarian aid, and damage descriptions. Positive emotions are associated mainly with social support topics, whereas negative emotions are closely linked to descriptions of property loss and casualties. The deep learning-based damage assessment model excels at identifying extreme damage levels, but semantic ambiguity at moderate levels reveals the subjective limitations of public expression. Compared with existing single-dimensional studies, the proposed three-dimensional framework comprehensively characterizes post earthquake social perception features and offers a replicable methodological pathway for social media-based public opinion monitoring and digitalized emergency decision-making.