iTunes Music Recommendations


The University of Illinois is working on a music rating project. You download an application that uploads your current iTunes library, and, some time later, it gives you a list of recommended songs. There’s plenty of info there, so I encourage you to rummage around. They even have a nice suggestion for applying ratings to unrated songs in your library. The server is a bit busy, so expect the result to take on the order of twelve or more hours before it’s ready.

Since picking up my new laptop a few months ago, I hadn’t bothered to transfer the actual ratings from my old laptop, just the tunes. So, I let it set the ratings based on the current play counts which, as of late, rather heavily favor The Allman Brothers’ latest, Hittin’ the Note. The results have been rather interesting. You can view the recommendations it gave me via this RSS feed.

I’ve since gone through and set the ratings for my entire library (all 880 songs), and have resent the library. The results won’t be ready for at least another ten hours, so the early birds will get a chance to see the difference between my current interests and all-time favorites. Also, if you’ve tried the service, let me know what you think.

Personally, I’m fascinated by the whole notion of “Collaborative Filtering Algorithms,” particularly with respect to things like music, because I’m interested in finding music that I like but a lot of it isn’t necessarily something that I’ll hear on the radio. Has anyone actually heard any songs from the Allman Brothers’ latest album on the radio? It’s a great album, but I’d never have found it if I hadn’t been an Allman Brothers fan, and caught the “FreeView” on DirecTV.

The question, though, is how do you test this kind of thing? If someone with fairly eclectic tastes, like mine, is satisfied with the result, does that really help? Or, as is actually the case, what does it mean when the algorithm returns a list of music with which I’m already familiar? I’m hoping this has potential to be something more than just someone’s intellectual exercise, but I’m not really sure.

&nbsp

Rick

Comments (6)

  1. matthew says:

    cool, so where’s the WMP version?

  2. steven says:

    how long did it take for you to rate your collection? and how do you deal with the "I like this song best when I’m in mood X, but don’t when I’m in mood Y, when I like that one better" delima?

  3. Joku says:

    This is a very interesting topic, making a algorithm that would find good music that way will no doubt not work for everyones tastes as I’ve found that my friends often find a track good for totally different reasons than I do. I have analyzed the tracks I most like and there’s indeed a pattern to what I like, even though the genres of music could be from any genre.

    Also I’m "one of those" who do not listen to particular artists much, instead most of the time I only like one song from particular artist. I only know a handful of artists which have say more than three good songs.

    I could talk about this for a year, but that would not be very productive.