When I first read about Hunch’s Twitter Predictor game, I was pretty skeptical. The game asks you to put in your Twitter user name and based on who you follow and who you are followed by, it predicts how you will answer questions on Hunch. Then I used it. It’s awesome. Well, pretty awesome.
Out of 35 questions I answer, Hunch correctly predicted by my answer to 32 of them and was only wrong with 3, 91% correct. And these aren’t just “yes” or “no” question, some have several possible answers. In fact, the game got so many right that at first I was sure it was all fake and they were just saying they were going to pick what I eventually did. Then I noticed the “take a peek” link, which tells you before you answer the question how you’re going to answer it.
I also wondered if Hunch was simply predicting how I’d answer based on other Hunch questions I had answered on my account. But actually, the game works even if you’re logged out of your Hunch account.
So yes, the predictor made by new Hunch employee Ben Gleitzman (a former Googler) is very accurate. But then I noticed something. As I played it again in another browser, the game asked the exact same questions. And the first question is always about my age range. So this is likely one of the keys to how the predictor works. Another friend had a series of questions that made it clear she was a woman — likely another key predictor.
I would bet the game is quickly scanning your Twitter followers and getting some obvious topical data, such as age range and sex. Then it uses the aggregate Hunch data that the service has collected over the past several months.
Still, it’s a pretty cool idea. And a great way to show off the data Hunch is collecting. The team answers more about the game here.
Hunch is a consumer web application that is building the “taste graph” of the internet, mapping every person on the internet to every entity on the internet and their affinity for that entity. An entity could be a web site, a cookbook, a hotel room, a celebrity, a restaurant, etc. Hunch creates a taste profile by asking them a series of questions which range from serious to profound and subsequently can make recommendations personalized to that user, which live in...