3 ms·

Except that those recommendations are really poor. And it's not clear to me at all that there's any way to scale the incoming dataset large enough to get useful

by jfb 11y ago

Except that those recommendations are really poor. And it's not clear to me at all that there's any way to scale the incoming dataset large enough to get useful results from it. Netflix might have good algorithms but they are handicapped by the shape of their catalog. Music has a much, much larger selection space and currently nobody has a sufficiently deep usage history to make "if you listened to x, you'll likely listen to y" work. Anywhere but smack dab in the center of the distribution algorithmic music recommendation engines fail completely (save for Pandora, which has a different small catalog problem.)

dperfect 11y ago
I agree, especially about Netflix and Pandora's poor recommendations as a product of their limited catalog. I don't however think that humans are any better in this regard. Humans also tend to recommend near the center of a distribution of tastes, and when they stray from that, it's out of a deliberate move to find something new/different. Algorithms can do the same (and many do), but many of those suffer from other problems - such as poor interfaces or limited catalogs as mentioned. Just because no one has done it really well yet, that doesn't mean we should all go back to human-curated playlists in my opinion. Edit to add: Perhaps part of my reluctance to rely on DJ's and human-curated playlists is a lack of trust. When a person recommends something to the masses, my gut instinct is to question motives (who is being promoted and why). When an algorithm recommends something to me as part of an unlimited service, I tend to feel like it's been tailored to me for no reason other than to maximize my enjoyment (though admittedly, the truth is that algorithms may be built with just as much - if not more - questionable bias as any human).
- jfb 11y ago
  I get my music recommendations from music blogs or my local record store, and I always always always find them more relevant than anything an algorithm poops out. Finding a way to scale that up would be interesting, but I also don't really think that there's a lot of money in it. Google could do this, really, or Facebook, I suppose, because they have the huge amount of activity data that could be used to take the place of "Hey, I like this Galactic Cannibal record, what else should I be listening to?"

by jfb 11y ago

dperfect 11y ago

I agree, especially about Netflix and Pandora's poor recommendations as a product of their limited catalog. I don't however think that humans are any better in this regard. Humans also tend to recommend near the center of a distribution of tastes, and when they stray from that, it's out of a deliberate move to find something new/different. Algorithms can do the same (and many do), but many of those suffer from other problems - such as poor interfaces or limited catalogs as mentioned. Just because no one has done it really well yet, that doesn't mean we should all go back to human-curated playlists in my opinion. Edit to add: Perhaps part of my reluctance to rely on DJ's and human-curated playlists is a lack of trust. When a person recommends something to the masses, my gut instinct is to question motives (who is being promoted and why). When an algorithm recommends something to me as part of an unlimited service, I tend to feel like it's been tailored to me for no reason other than to maximize my enjoyment (though admittedly, the truth is that algorithms may be built with just as much - if not more - questionable bias as any human).

jfb 11y ago
I get my music recommendations from music blogs or my local record store, and I always always always find them more relevant than anything an algorithm poops out. Finding a way to scale that up would be interesting, but I also don't really think that there's a lot of money in it. Google could do this, really, or Facebook, I suppose, because they have the huge amount of activity data that could be used to take the place of "Hey, I like this Galactic Cannibal record, what else should I be listening to?"