If I hit the Discover tab on my Spotify page it lists "Top Recommendations" for me.
On the top line at the moment (above) are Mice Parade, Markley, Yusuf Lateef, Can, Bill Frisell, ESKA, Wizz Jones, James Yorkston, Les Ambassadeurs Du Motel De Bamako, David Lang, Quentin Sirjacq, Greg Foat, M. Ward, John Fahey and Land Of Kush.
I've heard of only seven of those names and of those I have consciously listened to just four.
Further down the same page Spotify lists albums by artists it thinks I might like based, more specifically, on my listening to Max Richter, Judee Sill, Jamie X, Led Zeppelin and the English Chamber Orchestra.
Then there are some people I might like based on my listening to people I'm not aware of having listened to at all, artists such as Tsegue-Maryam Guebro and Akira Kosemura.
Further down the same page Spotify lists albums by artists it thinks I might like based, more specifically, on my listening to Max Richter, Judee Sill, Jamie X, Led Zeppelin and the English Chamber Orchestra.
Then there are some people I might like based on my listening to people I'm not aware of having listened to at all, artists such as Tsegue-Maryam Guebro and Akira Kosemura.
If I scroll down the entire page, which contains over 300 recommendations, I would guess I have heard about twenty of them. And I've heard a lot of music.
I'm assuming that somewhere in Spotify is a digital genie, tracking the things I listen to, noting the ones that I switch away from as well as the ones I return to, until it builds up some cloud-like picture of my tolerance. In which case it seems to be doing quite a good job.
Previous experience with recommendation engines has tended to result in a menu of records either so obviously in my taste cloud that I could have thought of them myself or so popular that the only explanation for why I hadn't chosen them myself was prejudice (which is, let's not forget, always a huge part of what people do and don't listen to.) The introduction of science into the serendipitous business of musical taste has tended in the past to come up with results too broad to tempt anyone. I might like Joni Mitchell. No kidding. These Spotify results have avoided that by proffering stuff I haven't heard and taking cues from the things I listen to on Spotify, as opposed to the far broader church of things I happen to like.
Apple's competing foray into the world of streaming, announced last week, makes much of their investment in "curation" by real DJs like Zane Lowe. I'm not sure about this. Disc jockeys are the very last people I would seek recommendations from. That's because they are required by their professional code and inclined by their temperaments to pretend that there's lots of new and exciting music coming along every single week. There isn't.
I'm assuming that somewhere in Spotify is a digital genie, tracking the things I listen to, noting the ones that I switch away from as well as the ones I return to, until it builds up some cloud-like picture of my tolerance. In which case it seems to be doing quite a good job.
Previous experience with recommendation engines has tended to result in a menu of records either so obviously in my taste cloud that I could have thought of them myself or so popular that the only explanation for why I hadn't chosen them myself was prejudice (which is, let's not forget, always a huge part of what people do and don't listen to.) The introduction of science into the serendipitous business of musical taste has tended in the past to come up with results too broad to tempt anyone. I might like Joni Mitchell. No kidding. These Spotify results have avoided that by proffering stuff I haven't heard and taking cues from the things I listen to on Spotify, as opposed to the far broader church of things I happen to like.
Apple's competing foray into the world of streaming, announced last week, makes much of their investment in "curation" by real DJs like Zane Lowe. I'm not sure about this. Disc jockeys are the very last people I would seek recommendations from. That's because they are required by their professional code and inclined by their temperaments to pretend that there's lots of new and exciting music coming along every single week. There isn't.
The music business, together with that part of the music media that aligns itself so closely with the business that it may as well be part of it, spends all of its energies on the new, new thing. The public, on the other hand, doesn't.
My problem with the automatic reccomendations is that they usually seem to think that if I for example listen to 20 Teenage Fanclub songs and one by the Rolling Stones, then I must want to listen to much more Rolling Stones, just because that is what a lot of other people do.
ReplyDeleteI once sought out a Bay City Rollers track on Spotify for research purposes, guess what Spotify then assumed I like and still like.
ReplyDeleteThe same goes for eBay and Amazon. I once bought an Arsenal loving friend of mine an Arsenal book on Amazon, cue endless recommendations for every Arsenal book ever written, which, as a Spurs fan, is a complete anathema to me. I won't be doing that again.
Interesting thoughts here.
ReplyDeleteNo matter how good the recommendation algorithm, it will soon annoy if it won't listen to your reaction - like Peter's Amazon Gooner-feed. A simple half-second "did you like this recommendation" yes/no box doesn't irk the user anything like as much as a constant diet of wrongness. Otherwise it's not adapting to the user's tastes and is instead acting as an unlistening broadcast medium pumping out best-guess adverts.
I'd be very happy for folks such as the 6Music DJs (esp. Cerys, Guy Garvey, Green Gartside) to do some playlist curation. They don't just chase the new, but find stuff that's fresh to my ears or that sounds fresh in the context they create.
ReplyDeletePS: My Spotify reckons I should check out this Ed Sheeran guy.
Consumption of music is active and passive but the numbers of individuals actively participating at a high level is a tiny fraction of the whole.
ReplyDeleteMost people are happy to be 'served' their music with little interaction and, as usual for a major corporation, Apple are serving the mass rather than the individuals like you. We actually need more 'educated' or informed arbiters to guide us through the mass of stuff that is released.
The only reason it won't be successful is that people won't easily break existing habits.
that took a minute.
ReplyDeletethe final argument was hard to grasp.
i think you're saying that DJs play "new" tracks that end up sounding a lot like things you've heard before and may not want to hear again, while ignoring old tracks which you might actually like but don't get played because they aren't part of any current profiteering scheme.
No, I'm saying DJs are required by their profession to pretend that new stuff is exciting per se.
ReplyDelete