Tom Conrad, CTO of Pandora, writes in his blog that starting tonight Pandora will start to incorporate community feedback into their system. Tom admits that sometimes the content-based approach that Pandora has used up until now 'just got it terribly wrong'. So to fix this, Pandora will  use the listener ratings to augment their content-based recommendations. 

It will be very interesting to see where this goes. The content-based recommender approach that Pandora has taken up until now can avoid many of the pitfalls that plague collaborative filtering systems.  Up until now, Pandora has been immune to popularity bias, recommender feedback loops and inertia that can plague systems like last.fm. (Just look at how many months the song 'Such Great Heights'  been #1 or #2 at last.fm).  Now as Pandora starts to mix in user data, they become more like last.fm, but without all of the last.fm goodness (no social browsing, no charts, no web services api, no song tagging).   It seems to me that if you are going to 'go social' you need to do it all the way.  Let me look at the Pandora charts to see what people are listening to. Let me find other listeners with similar listening tastes and see what they are listening to.   - Thanks Jeremy for the tip!

Comments:

Post a Comment:
Comments are closed for this entry.

This blog copyright 2010 by plamere