Associated Factors of Depressive Expression in Social Media During Emergencies: An LLM-Based Analysis.

Emergencies not only affect economic development and social stability but also pose serious challenges to the physical and mental health of the public. Depression is one of the most common chronic mental disorders following such events. However, few studies have explored the factors associated with depressive expression during emergencies. This research focuses on three core dimensions: cognitive appraisal, emotional response, and individual traits. It examines how perceived susceptibility, perceived severity, anxiety, sadness, self-focused attention and psychological resilience are associated with depressive expression levels. Unlike conventional approaches, this study employs a fine-tuned large language model (LLM) to detect depressive expression from social media data and aggregates the identified signals into hourly indicators for longitudinal regression analysis. Built on ChatGLM3 and manually annotated Weibo posts, the fine-tuned LLM has shown strong performance in Chinese social media depression recognition. The results show that these variables are significantly associated with depressive expression. Specifically, perceived susceptibility, perceived severity, anxiety, sadness, and self-focused attention function as risk factors, while psychological resilience serves as a protective factor. Our findings deepen the understanding of factors associated with depressive expression during crises and offer theoretical and empirical foundations for dynamic mental health monitoring and timely intervention.