Hunch co-founder Chris Dixon released a cool mashup called Forage Thursday night that utilizes the Hunch API to build personalized YouTube (s GOOG) playlists based on the people you follow on Twitter. Just enter your Twitter username on Forage.com, select a music genre, and the site will generate a playlist with 20 YouTube music videos. You don’t need to have a Hunch account to try the service, but it’s nonetheless a very neat demonstration of the power of Hunch.
So how does it work? Dixon told me via email that Hunch uses preferences from existing users to predict recommendations for users not known to the system. “Basically we have about one million people who have Twitter (or) Facebook connected into Hunch and told us about their music preferences,” he said. “We use this ‘known’ data to make inferences about unknown users based on who they follow, who follows them, and other signals.”
Dixon also emphasized that Forage is exploratory by nature — it won’t necessarily list your favorite bands, but rather music you may want to try. And I gotta say, it actually works pretty well for that purpose. Not only does it serve up an interesting selection of videos, users can also link to the resulting playlists. Unfortunately, it struggles with one of the issues many music-focused YouTube mashups are facing: Some major labels aren’t allowing third party sites to embed their clips.
Don’t expect Dixon to invest a lot of time into resolving this issue. “Forage is really just a little hack I did,” he told me said, adding: “And I’m not a great programmer.” Still, it’s an interesting little demo of the power of Hunch and its take on taste-based predictions, which in turn could be really useful for many online video platforms and services.
Recent research from the BBC has shown that content recommendations have a significant impact on the time people spend watching video online, but companies like Netflix (s NFLX) have in the past struggled to add social elements to these recommendations. Hunch’s approach of using known patterns to predict unknown preferences could help to make these types of recommendations much more valuable.
Chris Dixon stopped by our office a few months ago to talk about his experience as an angel investor and his thoughts on the VC model. Check out the first part of the interview embedded below, and read this post for the second part: