Goombah music discovery service

Goombah is a music discovery service that learns what you like, hooks you up with other people whose music tastes are most similar to yours, and introduces you to new music.

Goombah is a free service that analyzes your iTunes library and connects you to people and music that closely match your taste. Get recommendations based on your playlists or any mix of tracks, artists and genres. Goombah recommends  music from well known artists, and undiscovered emerging artists...all tailored to your musical taste.

You can download the Goombah application free at www.goombah.com

As most of you know, I was VP of product development at Napster. I love music and technology so Goombah is lots of fun for me. I also know the CEO of Goombah and want to help her get some recognition. Try it out and let me know what you think.

Subscribe - To get an automatic feed of all future posts subscribe here, or to receive them via email go here and enter your email address in the box in the right column.

Comments

[Anonymous]

This is an idea I had a long time ago (pre-itunes, WinAMP 2.x days), but I'm working on something else so I never spun any cycles on it.

It's a neat idea, but it seems more like a feature than a company. This is something a 25yr old kid (me) could write in his spare time, complete with a MySpace widget. Trend analysis, swarm heuristics, fuzzy logic, Bayesian probability algorithms can all be written in a night - data is data.

For a company to be built around this screams bubble to me.

Here's the piece they're missing: http://www.xspf.org/

[Andrew]

Don, I applaud your giving airtime to a service aimed at those who have adopted a rival for Microsoft's Zune. I refer, of course, to rivalry for the digital restrictions monstrosity (DRM) award...

[Henry G]

I tried out Goombah a bit and I am not very satisfied. My music library is so diverse that for the program to make accurate recommendations it would have to at least take into consideration song ratings and the number of plays. Furthermore, I think that I would simply prefer to use Pandora's box to discover and listen to new music.

[Gary Robinson]

To Henry G:

Hi, I'm Gary Robinson, CTO of Emergent Music LLC which makes Goombah. Thought I'd respond to your message.

Goombah DOES use play counts. A particularly neat thing about integrating with iTunes is that it also gets to use the counts from an iPod -- they get updated in iTunes when the iPod is plugged in.

We don't use ratings. The reason for that is that people use ratings different ways. For me personally, I assign songs I love but have heard a zillion times a low rating to reduce the probability of their coming up in shuffle mode. We have been considering adding a feature where ratings are OPTIONALLY considered and I think we will do that.

With regard to recommendation quality: we are just beginning to "push" our product so that people can know about it. The number of users we have is still relatively small. Our methodology depends on being able to match you with users who have similar tastes to you -- if there aren't such people (or only a couple) in our database, it does hurt our ability to give you good recommendations.

This is an effect that will disappear as our user base grows. I fully expect it to work very well for everyone -- even you -- in the not-distant future. Also, as we get more free music into our system, there will be more free music that matches your tastes.

Frankly, I am unexpectedly pleased with how well it's working with the number of users we have now. A lot of people are already loving it. For instance, here's a comment we received last week:

"You guys are costing me money, but that's a good thing! Wow I'm finding loads of great music, but there's already been a free song or two that influenced me to buy entire albums. Well so far I bought Jupiter Rising's self titled album on iTunes because of the freebie "Go!". I'll also be buying Michelle Albano's other albums ..." (it goes on in that vein).

And I must say it's working really well for me personally too.

So that it doesn't seem I'm just blowing smoke when I say that this *will* work when we get to critical mass, I'll mention two things.

1) We use a grid "massively distributed processing" model where we have far more processing power to apply to finding matching members and generating recommendations than any of our competitors do. To reliably find the very best matches is very computationally expensive. There are certain ways to do it on a small number of server CPU's, but they require summarizing the data in ways that diminishes accuracy. It's also possible to do it on a huge number of CPU's, which requires much more overhead and affects how you run your business. Google, of course, is able to do that for searches by selling targeted ads on the side, but our computational load is far heavier than theirs when adjusted for the number of users (again, to the extent that we don't want to compromise accuracy by reducing the data through a summarization process). We think that the grid approach is the right approach for the music problem.

2) The mathematics has been under development for years and years. A side story may be helpful. At the time that email spam was beginning to be a problem, I was one of those who was quite annoyed by it. Enough that I wondered whether some of the the "music filtering" math I'd been developing could be used for "spam filtering." I wrote an article for Linux Journal that used some of the math techniques and they were subsequently picked up by the anti-spam community and subjected to competitive testing against other techniques the community was considering at the time. The end results is that they were incorporated into a number of award-winning spam filters including SpamAssassin and SpamBayes (which won Editor's Choice awards from PC mags) and SpamSieve (which won MacWorld's Software of the Year award for 2003). But again, that was just a spinoff of techniques that were always focused on the music recommendation; the techniques in Goombah are very finely tuned for our purpose.

If you know other people with tastes that are similar to yours, I would (of course!) recommend that you get them into Goombah so that Goombah can quickly get to the point that it works well for everyone -- although the evidence I've seen is that it works great for the majority of people now.

If you have any other comments or feedback please let me know.