Multi-Model Fusion-Based Hierarchical Extraction for Chinese Epidemic Event

Area of Science:

• Computational linguistics
• Epidemiology
• Public health informatics

Background:

• Coronavirus disease 2019 (COVID-19) poses a global health challenge, necessitating robust surveillance systems.
• Extracting structured information from epidemic case reports is crucial for effective outbreak control.

Purpose of the Study:

• To develop and evaluate a novel method for Chinese epidemic event extraction (EE) from COVID-19 case reports.
• To improve the accuracy and efficiency of identifying epidemic-related events and their arguments.

Main Methods:

• Definition of epidemic-related event types and argument roles.
• Manual annotation of a Chinese COVID-19 Case Report (CCR) dataset.
• Proposal of a multi-model fusion-based hierarchical event extraction (MFHEE) architecture.

Main Results:

• The proposed MFHEE architecture demonstrates superior performance in extracting epidemic events on the CCR dataset compared to baseline methods.
• Experimental results on generic datasets indicate good scalability and portability of the MFHEE model.
• Ablation studies confirm the significant contributions of the hierarchical structure and multi-model fusion strategy to model precision.

Conclusions:

• The MFHEE model offers an effective solution for Chinese epidemic event extraction, enhancing COVID-19 surveillance capabilities.
• The multi-model fusion strategy addresses recognition bias, leading to more accurate event extraction.
• The developed dataset and model provide valuable resources for advancing research in epidemic intelligence.