Your next tweet could help track a Salmonella outbreak to a restaurant source if a new computer program from researchers at the University of Rochester enters common use. The system, called nEmesis, combines machine-learning and crowdsourcing techniques to analyze millions of tweets to find those from restaurant patrons discussing foodborne illness after eating at a particular location.
In its four-month pilot, the system collected some 3.8 million tweets and traced 23,000 restaurant visitors in New York City, finding 480 reports indicating likely food poisoning. According to the scientists, led by Henry Kautz, chair of Rochester’s computer science department, their findings jibe closely with public health department inspection data. Their findings will be presented at the Conference on Human Computation & Crowdsourcing in Palm Springs, Calif., in November.
The problem of foodborne illness is almost tailor-made for using linguistic patterns as a method of detection, according to Vincent Silenzio, MD, associate professor of psychiatry and a co-author of the paper. “People really do overshare!” he says. “We often roll our eyes when someone shares a photo of the meal they’re eating or the restaurant they’re in online, but in reality, they’re providing an enormous, non-invasive remote sensing system for picking up patterns in the information that has a real world application.”
The program uses an algorithm “trained” to spot tweets that are likely to be related to foodborne illness and linked to a particular restaurant. “We had to create a modeling environment that could suggest not only whether someone is complaining about illness after eating at a restaurant, but also that they attended that restaurant in the clinically relevant recent past,” says Silenzio.
What do restaurant operators think of the program? They haven’t had much of a chance to review it, says Katie Laning Neibaum, director of advocacy communications for the National Restaurant Association. “We would really need to take a deeper look at the system.”
Silenzio notes that the program would be a foodborne illness lawyer’s dream. “But that is complicated by legal and liability issues because it doesn’t amount to more than hearsay. But I think it would be very useful to the public health system as a decision support program or at least a corollary to their public health investigation systems.”