Researchers Mine Old News To Predict Future Events

hypnosec writes "Microsoft Research has teamed up with the Technion-Israel Institute of Technology to develop software that can predict events like outbreaks of disease or violence by mining data from old news and the web. The project, if successful, will result into a tool that would provide information that is more than just educated guesses or intuition. The team consisting of Eric Horvitz from Microsoft Research and Kira Radinsky from Technion-Israel Institute tested the program with articles from New York Times spanning over 20 years from 1986-2007."

Re:Past performance...

[Trepidity]

True, though it can to the extent that there are recurring patterns and you find the right ones.

On the other hand, there's also a circularity problem. Say you find, from analyzing 20 years of correlations, that certain events tend to happen some period after certain news reports. This might impact whether that relationship continues to hold in the future. That's already quite common for financial events: if you can reliably predict that when News Report Type X happens (for a possibly complex "X"), then Stock Move Y will happen, you can profit from it, but only until it becomes known by enough people, after which the arbitrage opportunity will close.

Re:Asimov was here

[gmuslera]

More important than the mythical Mule (unless telepathy really exists and to that extent), the important message in Asimov's Psychohistory is that predicted people shouldnt be aware of the predictions on them, and that includes the government. Is a good way to invalidate predictions, acting with the knowledge of the prediction instead of acting "naturally", whatever be it.