I’ve been playing with Goombah this weekend, and though it’s not perfect (what beta is?) it’s very intriguing. It’s an application that catalogues your iTunes library and feeds it into a database containing the libraries of other users. Goombah looks for matches between your playlist and other users, analyzing similarities. It then generates a ranked relevancy list.
It also creates a list of other Goombah users based on musical compability. You can drill down into their playlists to sess what they’ve got. Finally, and perhaps best of all, the service has a few hundred free tracks that are placed at the top of the list for your consideration. Every “Free Music Friday,” Goombah adds new tracks, some of which are from reasonably big names like Ani DiFranco, Sufjan Stevens, Calexico, XTC and Spoon.
Goombah is powered by BitTorrent, at least in terms of how it gathers user data, but it runs on trust. By standing between the
record companies artists and music fans, it acts as a taste broker. That only works if Goombah’s user community is confident that the program isn’t gathering data for piracy lawsuits, and artists are certain that it’s spreading their music without depriving them of a chance to earn a living from it.
With that understanding in place, there’s a real opportunity to build a new and different musical order that levels the field and allows all artists unprecedented reach to their fans.
Of course, the algorithm has to work, and I’m not so sure at the moment. I have over 10,000 tracks in my library, ranging from vintage blues to metal, with pop, folk, country and world music situated in between. Why, then, did the free tracks (the new stuff that I likely haven’t heard before) situate on such a narrow range of styles? I glanced at the entirety of Goombah’s free library, and there were several other songs that might have been recommended based on my tastes.
It’s unclear which metadata is used for the relevancy matches. Does the program check play count, last played, play order? There’s also a basic problem with ID3 in iTunes – if a track is skipped before it ends, it’s not counted as having been played. With live albums that’s a real issue, since many tracks end with the long spoken introduction to the next song. I tend to jump to the end of those. As far as iTunes is concerned, I never played them.
It doesn’t look like Goombah does ranking based on iPod data, but if play count isn’t factored in, that’s probably irrelevant.
Most iTunes metadata is provided by CDDB, which can be problematic. For example, CDDB classifies Ellis Paul in the “Soulful Rock” genre when he’s pretty much a folk artist. Blank fields present a unique issue. There are more blank “genre” tracks in my library than I can count. Add in tracks traded on Limewire and the like, with bad or no data, and it gets worse.
These aren’t Goombah’s problems, but unless these challenges are addressed, it will affect the usefulness of the program.
These problems are in the current Goombah release (0.9661).
Here’s a specific issue, which I’ve reported to Goombah. When I first fired the program up, the “free track” list included five recommended artists, with an “adventurousness” scale set near the middle. When I increased that level to “high” the artist count DROPPED to one. Further, the remaining name was conicidentally a “featured artist” on Goombah’s home page.
On the other hand, the non-free list is packed with cool looking suggestions. My top ranked user match had a Mindy Smith record listed that won’t be released for another six weeks. What a cool way to get a news flash.
One note: I’m running both the PC and Mac (Panther) versions. The aforementioned problem only occurred my Win box (XP SP2).