Transparent, Steerable recommendations - the video
We made a video of our Music Explaura - a web application that gives
you transparent, steerable recommendations. This demo shows how we can
start from an artist that we like (Jimi Hendrix), and steer the
recommender toward the aspects of Jimi Hendrix that we like (his guitar
playing) and away from aspects that we don't like as much (60s
psychedelia). The resulting recommendations are much closer to our taste
than the typical recommendations. Watch the video:
You can read more about transparent recommendations in this two pager: Creating Transparent, Steerable Recommendations as well as this blog post.
I love the fact that you accurately call them "The Cream." Looks cool, when can we all goof around with it?
Posted by Zac on November 03, 2008 at 02:17 PM EST #
Looks really interesting (!), especially the feature where one can drag the tag which one is most interested in - and thus make a 'weighted' search. Just one question though; where does the 'SITM' project itself (analyzing the various features of a track such as timbre or energy) come in? As far as i understand, the music explaura is all about using tags gathered from the web? Is there a clear connection between the music explaura and the part of the SITM project which deals with anayzing music?
Posted by Stian on November 03, 2008 at 02:49 PM EST #
Stian:
In parallel to the text-based, steerable recommendations, we've been building a music autotagger. The autotagger can learn how to predict what social tags will be applied to music - so if we have a brand new track, a track with no tags - the autotagger, using only the audio, can apply a set of weighted tags to the track. These autotags helps us deal with the cold-start problem - we don't have to wait for hundreds or thousands of listeners to tag a track before we can start to make recommendations for it, we can autotag the track and incorporate it into the recommender right away. You can read more about the autotagging technology in this paper: http://www.iro.umontreal.ca/~eckdoug/papers/2008_jnmr.pdf
Posted by Paul Lamere on November 03, 2008 at 02:59 PM EST #