the company uses advanced search algorithms that can pull the correct data from articles and other written content around the Web. It’s called open information extraction, something Etzioni has studied with colleagues at the UW.
“The key is that a lot of that information is in textual form, which makes it challenging to extract and to place in the right spot,” Etzioni says. “A computer can’t read. So how do we know that particular article was in fact highly relevant to that Canon Powershot?”
By using open information extraction, Decide can mine the written materials—blog posts, press releases, and so on—that contain crucial information about upcoming product releases, and add that information to its recommendations about whether a user should buy now or wait a while.
These kinds of questions are being answered in some places on the Web right now, Etzioni says. “Even Google, if you type in “release date Madden 12,” you’ll see at the top they’ll actually extract the date. They won’t just match the words. So people are doing it around the periphery,” he says. But research at the UW has prompted Etizoni to prod the industry toward more action: “Let’s just go whole hog and do it for everything.”
The university itself is hoping to nudge things along by releasing an open-source version of the open information extraction software ReVerb. The commercial corollary is Decide. “We’re really raising the bar substantially in terms of using information extraction on shopping,” Etzioni says.
There are other people working on improved electronics shopping search, including Shopobot and Retrevo. Google also recently bought Seattle entrepreneur Dan Shapiro’s startup Sparkbuy, which helped sort through laptops online. But none of those are playing in the more advanced area of price predictions.
So, years after Farecast’s success, why hadn’t someone else jumped on price-predicting electronics shopping first? Etzioni doesn’t quite know, and wonders why others aren’t taking the technology to all kinds of other consumer goods already. Zillow (NASDAQ: [[ticker:Z]]) takes this approach to home prices, he says, and SeatGeek employs predictions for tickets. But Etzioni says that’s about all he can think of.
“Very few people have said ‘This predictive pricing this is exciting, let’s do it for 50,000 different things,'” Etzioni says. “What about for cars? What about for apparel? I really think it’s a failure of imagination.”
Not that it’s a cakewalk. There are many terabytes of data to collect and organize before predictions can start taking shape—Decide itself was working on this task for a couple of years before the company really got rolling. “I don’t think it’s easy to do that piece of predictive pricing,” Etzioni says. “But I think we’ve paved the way, and I’m disappointed that more people haven’t done it.”
