last.fm vs. pandora
A long time ago I moved my music collection onto an external hard drive because it was taking up too much space on the internal hard drive. This wasn’t a big deal because I used my laptop pretty much as a desktop, leaving it plugged in and sitting on my desk all day. But when I moved to New York last year I found myself toting the laptop with me everywhere I went, which means I am disconnected from my tunes during most of the time that I would like to be listening to them.
So I became a pretty avid fan of Pandora in the Fall of ‘05. I knew of last.fm but when I first tried them they were only aggregating your music tastes through their iTunes Audioscrobbler plugin and, since I don’t use iTunes that much, I didn’t pay too much attention to them, until a few weeks ago when I came across a post about last.fm on Fred Wilson’s blog when I was reading it before heading to the Union Square Ventures offices for a meeting.
I discovered that they are also streaming tunes, just like Pandora, except of course they are using collaborative filtering, rather than algorithmic pattern analysis, to recommend music to you.
So: which is better? I like the music I hear at last.fm. It tends to be somewhat less obscure than what comes up at Pandora. Pandora’s the only place I’ve listened where I heard an “ah-ha” song and said, “I’ve got to get that mp3,” so that’s a definite point or two for Pandora. A few uninformed musings on the pros/cons of collaborative and algorithmic filtering:
- Does it suffer from a convergence effect? Most people won’t have a great breadth of musical knowledge, so their selections are likely to cluster around some of the more popular songs, right? This popularity will accelerate, too, meaning the initial top selections are likely to get an early lead and cement their places at the top.
- Hype must affect collaborative filtering in a similar way. The hot new band is likely to get a lot of initial play, which means it will show up more often due to its hype but not necessarily its quality. Algorithmic filtering is pretty much inherently immune to hype. It’s almost anti-hype.
- On the other hand, algorithmic filtering is very unlikely to cross genres. I find this on Pandora a lot. If I give it too many seed songs, the resulting playlist tends to cluster around the same small set of obscure, similar-sounding songs. Collaborative filtering is more likely to give you music that is good and might push you out of your comfort level a bit.
Either way, what amounts to free, very-high-quality radio (at both Pandora and last.fm) seems like a very good thing.
Leave a Reply