Computer-based language analysis identifies public health threats

Analysis of the words that triage nurses use to report on symptoms when a patient first presents to a hospital emergency room is providing the basis for faster detection of infectious disease outbreaks and possible bioterrorism events in Australia.

At the University of Melbourne’s School of Computing and Information Systems, Professor Karin Verspoor has used computational linguistics to automate the analysis of triage reports recorded into emergency room computer systems.

Computational linguistics models the way natural language is used in computer-based human communication. It allows you to see the higher patterns, themes and relationships expressed in natural language through computer processing, Professor Verspoor says

Her work is a research partnership with the Defence Science and Technology (DST) Group, which has been concerned about delays in identifying major public health threats. Current monitoring systems rely on patients presenting to doctors and doctors requesting pathology tests — which can take days to complete — and then reporting the findings.

Professor Verspoor found that using computational linguistics to analyse the emergency room description of symptoms could create a faster, up-to-date snapshot of the health of the population.

We can roughly categorise symptoms and keep count of people affected, triggering an electronic alert when patterns of certain symptoms change, she says. Her approach has been validated and successfully trialled in Victorian hospitals, but has not yet been implemented nationally.

The partnership with the DST Group has also found that potential public health problems can be identified from the automated analysis of language use in social media. In this case, the impact of strong emotions on language use provides the basis to differentiate when people are discussing disease in an abstract sense and its occurrence in a more personal context.

Computational linguistics can also be applied to help coordinate the response to a health threat. This includes tools to monitor all communications among responding personnel presented visually, to help ensure the proper level of coordination.

More information

Professor Karin Verspoor

Telephone: +61 3 8344 4902



YouTube: Karin Verspoor Presentations

Presentation slides: Leveraging Text Classification Strategies for Clinical and Public Health Applications

Presentation slides: Towards early discovery of salient health threats: a social media emotion classification technique