AI app uses social media to spot public health outbreaks
Image: iStockphoto.com/Denys Prykhodov
Each year, 48 million people get sick, 128,000 are hospitalized, and 3,000 die of foodborne diseases, according to the Center for Disease Control.
And, in the age of social media, many are tweeting about it.
A new app called nEmesis uses AI and natural language processing to pinpoint tweets related to food-poisoning, using geotagging to trace the outbreak to its source. The app was used by the Las Vegas health department to help inspectors pinpoint outbreaks.
The app, developed at the University of Rochester, was funded by the National Science Foundation, the National Institutes of Health, and the Intel Science and Technology Center for Pervasive Computing. Still in beta mode, it was presented at the Association for the Advancement of Artificial Intelligence (AAAI) conference in Phoenix, Arizona last month, where it won an Innovative Applications of Artificial Intelligence award.
“This is an example where machine learning applied to large amounts of data benefits public health,” said Adam Sadilek, a researcher on the project who is currently at Google Research. “It is one of the unexpected benefits of public social data analyzed in aggregate.”
The app was used in Las Vegas over a three month period, during which it monitored about 16,000 tweets from 3,600 users every day. A thousand of the total tweets were connected to a specific restaurant, and about 12 of those included information that pointed towards food-poisoning. Inspectors used the information to make a list of restaurants with the highest likelihood of violations. Unlike the previous system, which led to 9% citations of health violations, the AI-based system resulted in citations for health violations in 15% of the inspections.
On top of spotting outbreaks, the app also has the potential to give voice to the voiceless.
“Social media provides a new method for organizations that care about social good to ‘listen’ to the public at large, including people who are ignored by traditional channels,” said Professor Henry Kautz, director of the Institute for Data Science at the University of Rochester.
Kautz said that a grant from the CDC will allow him to expand the project to other cities. And, in addition to foodborne illnesses, the system can be “modified for other disease conditions,” he said. “Consider a health agency that performs mosquito control. We could prioritize locations for mosquito control, by data mining tweets about mosquito bites and symptoms of mosquito-borne illness,” he said.
The app is not yet public. Researchers have intentionally restricted its use, proceeding with caution. “We wouldn’t want to libel restaurants that were perfectly clean, but someone got sick just by coincidence,” said Kautz.
But, when it does go public, there’s the potential to save many from preventable illnesses—in a whole range of areas related to health and the environment.
“We are always searching for ways to utilize technology to improve our data collection and evaluation methods,” said Matt Rhodes, deputy director of the Department of Public Health and Wellness in Louisville, Kentucky. “The methodology used is something that could certainly benefit how we receive information to drive service requests.”
Read this article –