Digg getting into collaborative filtering
It looks like Digg is getting into collaborative filtering.
Collaborative filtering is the process of taking lots of data, and being able to make recommendations from it. For example, on Foodio54 if you rate restaurants we will determine who you are similar to and what other restaurants you may want to try. For Digg they will of course be recommending stories, but the concept is the same.
One thing that will be an interesting challenge for Digg is that their ratings have only 2 possible values, “dugg” and “buried.” For Foodio54 (and Netflix, the king of collaborative filtering) we have 5 star ratings so if you rate something a 5 and I rate it a 4, we’re not the same, but the algorithm can see that we are somewhat similar. This concept is also helpful because if we recommend something a 2, but to you it was a 3, that’s not so bad, where as if Digg thinks you should Bury something, but you would really Digg it that could be bad. It would be like us recommending a 1 for something you think is a 5.
Also, just a reminder, the best way for Foodio54 to give you the best recommendation is to go through and rate at least 10 restaurants so we can get an idea of what you like, the more you rate the better the recommendations will be.
If you haven’t signed up with Foodio54, it’s free, and you can get started today.
Tags: collaborative filtering, digg, netflix, recommender system
July 1st, 2008 at 11:07 pm
You obviously aren’t very innovative in your use of collaborative filtering.
Digg doesn’t have simply two levels of rating.
Favorited
Dugg
Indifferent
Buried
If you factor in comment similarities, you could get even better results. I don’t know if Digg does this, I’m just saying that it’s not exactly as challenging as you think.
July 1st, 2008 at 11:56 pm
Actually if you read the Digg whitepaper or watch the video with Anton Kast it clearly shows that they only use diggs, buries, and the topic that the digg is in. The math of collaborative filtering is really amazing in what you can accomplish with what may seem like a minimal amount of data.
As to our innovation, we applaud Digg for being so open about how their engine works, but we’re not ready to give away how we do this quite yet. We use a different method than digg, but trust me when I tell you that a lot of work went into our recommendation algorithm. In fact, in terms of development, Foodio54 started as a recommendation engine before we decided to recommend restaurants, and more work has gone into that algorithm than any other aspect of our development combined.
My point is that with only values of 1 or 0, I believe that it will be difficult to be able to give good predictions. We are certainly excited to see how this goes though.
Mike