Feb. 4, 2013 at 5:59 PM ET
New software under development aims to stop history from repeating itself by using old news and related data to warn of pending trouble in time to take corrective action.
The system could, for example, help international aid agencies assess the likelihood of a cholera outbreak in time to treat a population with a limited-duration cholera vaccine, explained Kira Radinsky, a researcher at Technion-Israel Institute.
“We’ve engaged with experts in the public health field who study cholera and we’re hopeful that systems stemming from this research might be useful one day for early warning about this disease and others,” she told NBC News via email.
Radinsky is developing the software in collaboration with Eric Horvitz at Microsoft Research in Redmond, Wash., as part of their exploratory efforts on ways to predict the future based on past information.
The researchers built their predictive model based on more than 20 years of historical news data from the New York Times. It examines past events with similar outcomes.
The software, which is nameless for now, also incorporates related contextual information pulled from LinkedData, a project that finds connections between hundreds of resources such as Wikipedia.
The combination allows the software to extrapolate from news of a cholera outbreak in Angola, for example, to predict a similar outbreak in Rwanda.
“Rather than be limited to the data about the details of an African country where a sequence of events and outcomes occur, the system learns to consider ‘African country with certain level of GDP X, cities with population of Y, and geographical features Z,’” Radinsky explained.
When tested on historical data, the system’s warnings of disease outbreaks, violence, and other events were between 70 and 90 percent accurate, according to MIT’s Technology Review.
“We are continuing to refine the approach and one day can imagine making a descendent system available to international aid agencies and to the public in general,” Radinsky told NBC News.
— via MIT Technology Review