Using AI to Understand Public Health Emergencies Faster: Research

During disease outbreaks (such as COVID-19 or other infectious diseases), a huge amount of information is produced very quickly, including government announcements, news reports, clinical updates,

emergency response actions and situation reports. Much of this information is written as unstructured text. As these are not organised in databases, extracting this information manually is slow and inconsistent, and existing Named Entity Recognition (NER) research which was primarily developed for natural hazards does not address the domain-specific terminology or severe label imbalance found in infectious disease contexts.

This study “Scenario-Response Extraction for Public Health Emergencies: Addressing Imbalanced Entities with a Chinese NER Model” lead by Asst. Prof Michele Nguyen is a collaboration with Dalian University of Technology and it addresses these gaps by developing a domain-adapted NER framework. Researchers created the first Chinese dataset focused on infectious disease scenario–response analysis, design a Triple Loss function to better learn sparse and low-confidence labels, and propose a MacBERT–BiLSTM–CRF model enhanced with R-Drop.

Accurate extraction of scenario–response information is essential for timely and informed public health decision-making. The proposed framework improves recognition of rare but operationally important entities and automates a task that would otherwise require extensive manual effort.

The research potentially enables faster situational awareness, more reliable interpretation of outbreak reports, and better support for emergency management activities such as monitoring, planning, and coordination.

 

C-AIM Lead:

Michele NGUYEN Asst. Professor of Data Analytics Lee Kong Chian School of Medicine Nanyang Technological University

Related Research Domain: Public Health, Human-AI Interaction

Keywords: Scenario-Response, Chinese NER, Public Health, Loss Function, Imbalance