Development and testing of a public health emergency intelligence analysis system based on text analysis and NLP...

To tackle challenges including delayed information support and inefficient decision-making in public health emergency response, this study develops an intelligence analysis system for public health emergencies based on emergency information management theory from library and information science.

Using 1,026 text data items such as government reports and flow survey records from the COVID-19 epidemic in Shijiazhuang City (1,033 confirmed cases), multimodal analysis methods were integrated, including logistic regression, C5.0 decision tree, TransH-based knowledge graph, and chi-square test. The BIO tagging scheme was adopted with annotations performed by three epidemiology professionals, achieving an inter-annotator agreement (Kappa) of 0.78.

Key transmission sites were identified by chi-square test (χ ² = 87.32, p < 0.001). Risk factors such as advanced age (OR = 3.15) and village clinic visits (OR = 4.72) were identified through logistic regression. A case-place-time network was constructed using the TransH algorithm (accuracy 0.89). The C5.0 decision tree classified high-risk areas (AUC = 0.91), and Apriori association rules revealed patterns such as "wedding banquet → family gathering" (confidence 0.86). A Python-based system improved intelligence extraction efficiency by 47.8%.

The study successfully establishes an interdisciplinary framework integrating library informatics, epidemiology, and AI. It identifies churches and wedding banquets as key transmission nodes, and village clinics as amplifiers due to delays in identification and reporting. The developed software tool enhances response efficiency, supporting rapid contact tracing and control strategy formulation.