Music Recommender Systems and Discovering New Music

Pretty much all of the online sources for music offer suggestions for music that we might enjoy based on the music we are listening to or considering buying.  

These suggestions are driven by what are called 'recommender systems.'  My goal is to gain a thorough understanding of the current state of these systems and potentially help improve them.  

It's pretty easy to make a good guess about how each service is implementing these systems.  But, as I have begun learning about recommender systems.  I can see that there are lots of details that can make some better than others.  Coursera and the University of Minnesota offer a MOOC about recommender systems.  (check it out here.)

So what does it mean to say some systems are better than others?  Good Question!  

I know what I want - and it's not suggestions based on what is most popular even if the definition of popular is limited to my friends, some specific genre (don't get me started on "genre") or what people with similar listening habits enjoy.

Most recommender systems in use today are much more sophisticated than simply filtering and suggesting what is most popular.  They can be curated, expert based, social based, content based, use collaborative filtering or some combination of all of these.

Without knowing all the details about each particular system I am still confident in stating that (1) they are not all the same, (2) even in their current state they do a lot to expose people to music they might not have discovered on their own and (3) they are constantly being improved.

I found a slideshare from a RecSys forum in 2011.  It's a great presentation titled "Music Recommendation and Discovery Remastered."  Here's a link.  Lot's of very cool information and ideas.  I'm just not sure how much of these ideas are actually implemented in the music recommendation systems popular today.

The sites using recommendation systems that immediately pop into my mind are; Pandora, Spotify, iTunes, Amazon, Rdio, eMusic, Google Play and last.fm.  I think we even have to add YouTube to the list. 

Having said that I'd like to start compiling information about music recommender systems.  Everyone is welcome to help!  Here's a link to a public Google spreadsheet.  Any help with this is greatly appreciated - add to it or change it in anyway you think will make it better.  Perhaps this will evolve into a poll later.

Ultimately, I'd love to see a recommendation engine include more content based attributes and allow me to tailor the weighting of attributes to my particular desires any time I am using the system.