1.5 million THAY questions answered

Since we opened up our site on Friday inviting around 30,000 users (and growing daily), we’ve had over 1.5 million Teach Hunch About You (THAY) questions answered. This is a phenomenal amount of data, and we’re going to be able to do a lot of things we’ve been hoping to do, not least of which is finding interesting correlations. As we’ve been discussing with rdl, striatic, tfelice et al in the forums, most sites that do collaborative filtering do it in one subject area only, i.e. Netflix will look at the movies you like and recommend other movies, Amazon will look at the books you’ve bought and recommend other books, and so on. We think there’s likely some cross-topic correlations that may prove interesting.

Think different

If a person says “I like to fit in and be in tune with those around me” — they’re more likely to get the “switch to a Mac” result, but if they say they like to be perceived as unique and different to make their own mark — it’s correlated with sticking with a PC! Mac users are not really thinking different after all, but following the herd. (Or not thinking differently, as Kelly and I, the resident grammarians, would say.)

Some other interesting things: People who believe in alien abductions are more likely to blame Nancy Pelosi for the financial crisis, and people who eat more often in restaurants than at home read Paul Krugman’s blog Conscience of a Liberal. Pass the arugula please?

Fun with Academics & APIs

Michael Kearns is a good friend and advisor to Hunch. (After we’d been working for months and months on Hunch he said, “We have an old saying in machine learning circles: Don’t learn what you already know.” Genius. ) We’ve been brainstorming with him about interesting academic work we could do with Hunch aggregate data. One study we’d love to see is if there is some set of 20 questions that predicts 95% of facts about people. Some interesting questions: Are we less or more unique than we think we are? Can our tastes in cellphones be predicted more easily than our taste in rock bands? When we open up the APIs and let the aggregate (anonymized!) data be analyzed more interesting stuff will emerge.

New Features

We’re planning a bunch of new features using all this data. We’d love to be able to show you your similarity to groups of users, show you Hunches we think you’d like, (Never go dancing? You might like Kierkegaard!), show trends over time, fads of yesteryear, predictions of what people will like next year. We’re really looking forward to being able to launch some of this stuff.

Help us brainstorm about the possibilities in the forums. And thanks for all your contributions.