Pandora vs. Last.FM

This is a great post by Steve Krause discussing the nature vs. nurture of Pandora vs. Last.FM...two music recommendation sites I've touched on before.


Though it is merely a side note, my favorite comment is:



"...better algorithms are nice but better data is nicer."

I mentioned in the often-referenced Big Picture post that I was wary of pure folksonomy for this reason (discussed under the usability section). I certainly like the theory behind Last.FM's approach regarding software that collects data in addition to user-submitted tags (though it is a tad big-brother). My spin involved restricted options for meta data and open-ended tags for other areas. Ideally, software could search for typos and plurality issues, but be subject to a human editorial process.


Though I've already states that my personal preference is for clean data and algorithms based on genre (not popularity) a la Pandora, Krause's suggestion of somehow combining the two approaches reminded me of where I most often hear about new music...my friends. Not virtual "friends" that I'd allow to join my personal network on a social networking site, but people I know and trust. So why not allow users to limit the recommendation factors only to others that they place on their recommendation list. All other's opinions are eliminated, only the opinions of the flagged users are factored into the algorithm (of course, this can be turned off, allowing all social data to be considered).