During the 2008 Presidential campaign, John McCain accused Barack Obama of being “the guy who worries about the price of arugula,” a suggestion that Obama was an elitist. Many scoffed at the remark, but according to Hunch CEO and co-founder Chris Dixon, liberals do prefer arugula while conservatives opt for iceberg lettuce. The connection between lettuce preferences and political orientation is something that Hunch has uncovered through its taste graph and recommendation engine, something that Dixon describes as “the most sophisticated system ever built for predicting human preferences.”
On stage today at ReadWriteWeb’s 2WAY Summit, Dixon sat down with our own Marshall Kirkpatrick to talk about how Hunch has built its taste graph and how this sort of recommendation engine may shape the future of a more personalized Web.
How Hunch Knows What You Like
By asking just a few simple questions to users, the Hunch website is able to predict with pretty astonishing accuracy how they’ll answer other questions. After you “tell Hunch about yourself,” the startup’s recommendation engine is able to offer suggestions about other things you might like. Kirkpatrick pointed to a recent infographic on Dixon’s own blog, detailing the amount of data that the startup is working with: about 500 million people, 200 million items and 30 billion edges. That last figure is key, as it means that Hunch has what Dixon calls a “known preference” for 30 billion items.
Dixon explained a little bit of the technology behind Hunch and behind other recommendation engines. He noted that unlike some systems that rely on natural language processing – on what people say, for example – Hunch is able to glean quite a lot based on who you follow on Twitter and what you like on Facebook. So if you follow Barack Obama, you’re more likely to be a liberal. And you’re more likely to prefer arugula.
Dixon said that initially Hunch had thought about building its own dataset, but instead has tapped into what’s already on the Web, utilizing Facebook and Twitter authorization for example in order to identify some of these tastes.
Hunch Knows What You Like: Privacy Concern?
In light of concerns about Google and Facebook building facial recognition technology, Kirkpatrick asked how (or if ) we can protect people’s privacy when it comes to “taste recognition.” What are the implications of being able to tell so much about a person – their sexual orientation, their political orientation – just by their answering a few questions or linking their Twitter or Facebook accounts.
Dixon made it clear that despite Hunch’s ability to predict users’ tastes that the company would never sell that data. “We have never made a data deal,” said Dixon. Furthermore, people can only get predictive information about themselves.
What Dixon envisions for Hunch nonetheless is to be “the place you keep your taste profile.” That’s something that’s made a lot easier, in no small part, thanks to the Hunch API. Dixon talked about the possibilities of incorporating the Hunch taste graph into various applications, from better follow suggestions on Twitter to better hotel suggestions on a site like Kayak.
Dixon says that he believes that there seem to be two ways in which personalization will occur in the future – “either through shadowy cookies and things behind the scene” or by things that users control. And we want to be the people doing it.”
