One of the data mining holy grails must certainly be predicting a user’s taste in music. More general there are many problems where a company would want to know if a certain bit of information will interest you. Targeted advertising like Google Adsense or the Chitika ads at the bottom of this page, are just the tip of the iceberg. Netflix is offering $1,000,000 to any one who can improve their recommendation algorithm by 10%. That’s serious money.
But this post is about music matching. There are at least two ways to tackle the issue:
1 Cross referencing titles and artists between different users. One of the most used approaches involves creating a gigantic database of the music a user plays. Next, statistical analysis enters the picture, and out come the users that have matching tastes and the tracks they will both like.
Two interesting programs offer this functionality. iRate Radio is probably the lesser known of the two. It does require a little effort, because you have to vote on all tracks. While the user interface might be a little spartan, it is easy to use and doesn’t get in your way with ads. Check it out, well, once the server is back up.
Last.fm is widely known and has an enormous log of users and music. This has the advantage that it will give you many, many options, but from my experience it seems to only recommend the popular and big-name artists. I haven’t really been surprised by the “My Radio” playlist. I don’t know why this happens, maybe it’s the algorithm, but it might as well be related to the fact that they don’t have licensing deals with small players.
To finish this section, I’d like to mention Anywhere.FM, a new player in this scene. As with iRate, it seems to be offline right now, but it looked very promising this afternoon when I managed to launch the player. It has the most fancy interface of the lot and it makes it very easy to go exploring your neighborhood (people that match your taste). It requires a bit more manual interaction than the Last.fm station, but the interface makes it really easy.
2 Analyse the music itself. Another solution is to take a look at the problem from the other side. Start from a known good piece of music and go look for music that is similar in tempo, tone, mood, etc.
Pandora is the prime example, however it is no longer available in pretty much the entire world due to licensing and/or copyright issues. Using the Music Genome Project, it classifies music based on over 400 attributes. From the little information that is available, this is apparently a manual endeavour. Many music experts enter these attributes into a large database, from there, computers take over. I must say, back when I could still access Pandora, it worked rather well and the recommendations where pretty accurate.
Musicovery starts from similar premises. However it looks like it has only about 10 attributes and certainly fewer songs in the system. It seems to always revert to a small set of artists that are linked to Amazon’s affiliate links (just a hunch of why the set is limited). Or maybe my choice in music is just that limited. I suggest you find out for yourself.
And if you still need a recommendation:
