Anyone remember that TV show “Early Edition,” where the main character mysteriously received the following day’s newspaper every morning? That was pure fiction, but present day computer scientists are working on a real version.
A team from Microsoft Research and the Technion-Israel Institute of Technology recently created powerful software that mines decades of old newspaper articles along with online data to predict future disasters. Their prediction system was able to forecast significant numbers of deaths with 70 percent to 90 percent accuracy, Tom Simonite reported in MIT Technology Review.
The scientists, led by Microsoft Research co-director Eric Horvitz, fed New York Times articles from 1986 to 2007 into their system. They also mixed in crowd-sourcing and other data from the sites DBpedia, WordNet and OpenCyc. Mining that data brought patterns to light, such as droughts preceding cholera epidemics in Bangladesh during the 1970s and 1980s.
Then the scientists tested the pattern they uncovered on other data and found it was remarkably accurate in forecasting large numbers of deaths. Unlike past work in this area that focused on mining the past, the BBC pointed out that this has the potential to be used in real time to forecast future events. The methodology and algorithm descriptions can be found in their research paper (PDF).
Although the group doesn’t have current plans to commercialize the research, Horvitz told Simonite that a refined version of their system could eventually be used by government aid agencies and groups involved in disaster response. That could make for some incredibly useful government alerts: “Hey citizen, you might want to put some protective gear on right now.”
Credit: Jane M. Sawyer