The reason algorithms force this to happen is the same reason the algorithms force images to outpace other content - upvote velocity, or in this case, listening velocity. Shorter tracks rack up...
The reason algorithms force this to happen is the same reason the algorithms force images to outpace other content - upvote velocity, or in this case, listening velocity. Shorter tracks rack up more listens which immediately bias them in the data towards more plays. It's a simple math problem if Spotify cared to solve it. It is absolutely not what the consumers want. The consumers aren't even paying much attention. They just loop their faves or bounce on to a playlist curated by humans, not algorithms.
Algorithms are useless in this area, really. No algorithm can ever understand music because it has no emotional context to appreciate that music from in the first place. Human recommendations are better than algorithms when it comes to music. Spotify is counting numbers when they should be building a social network for their listeners to share music instead.
Legit - it's a big problem for me with Spotify because a lot of the time I'll like or add to a playlist a song from a band, while not actually liking their other stuff that much... but then I get...
Legit - it's a big problem for me with Spotify because a lot of the time I'll like or add to a playlist a song from a band, while not actually liking their other stuff that much... but then I get them shoved into any "random" playlist...
I seem to get stuck in "genre holes" on Spotify. Sometimes my recommendations kick ass and I find new favorites. Most of the time, however, I feel like I'm trapped by one sound and Spotify won't...
I seem to get stuck in "genre holes" on Spotify. Sometimes my recommendations kick ass and I find new favorites. Most of the time, however, I feel like I'm trapped by one sound and Spotify won't let me break out. My Discover Weekly seems designed to push me back to things I already like rather than introducing me to anything new.
As I type this, Jungle by Jimi Hendrix is number three on my Discover Weekly. Jungle (this version anyway) is an instrumental in the middle of Both Sides of the Sky, it's not a song I'd listen to on its own like, say, Red House. Furthermore, I've liked 17 Hendrix songs; pretty good chance I've heard Jungle once or twice. Why am I being recommended Hendrix anyway? I don't need Spotify to introduce me to Jimi Hendrix, I need Spotify to show me new stuff that I don't already know about.
I find what works best is to make a playlist of at least 20 tracks of stuff you like, then make a copy of that playlist, hit the copy with the 'enhance' feature, and check out the songs it...
I find what works best is to make a playlist of at least 20 tracks of stuff you like, then make a copy of that playlist, hit the copy with the 'enhance' feature, and check out the songs it auto-populated. This is kinda forcing it to make you a discovery weekly on demand, as if that playlist were your liked songs.
While the songs themselves may not be the best choices, the artists it throws at you are probably worth a quick brush of their top tracks. You can keep on repeating this process a couple of times, and it will eventually run out of fresh content to enhance with and start repeating the same artists. That's the time to create a new playlists of other tracks you like and try it again.
Another way is to track down other music hounds on Spotify that share your tastes and follow them. The genre playlists being auto-generated by the service itself aren't useless, but they aren't deep cuts either. Other user playlists don't have that problem. Where else but spotify can you find 2,177 songs that 'french people would play at a party or just with friends around' still collecting new tracks all the time.
It annoys me when artists disappear from their service. I really don't care that Burger Records had some serious issues with exploitation - they paid for it and it's none of my business. I do care that their entire catalog of garage rock and other oddities, encompassing dozens of artists, has vanished from the service as if every single artist was personally involved in the scandal that took their publisher down. :/
I think Last.fm sort of hit on one of the most interesting ideas for this: let users tag music. Then someone could choose to view/listen to a tag, and find new recommendations that many people...
I think Last.fm sort of hit on one of the most interesting ideas for this: let users tag music. Then someone could choose to view/listen to a tag, and find new recommendations that many people chose to tag with a certain keyword. It works well for niche genres, which conventional music metadata doesn't really specify. "Dance Music" isn't helpful when I'm looking for Tech House or 90s Eurodance.
The peak for playlists for me was early 2010s 8tracks, which worked because of tags like you mentioned. I discovered all sorts of genres on there that I wouldn't have come across otherwise -...
The peak for playlists for me was early 2010s 8tracks, which worked because of tags like you mentioned. I discovered all sorts of genres on there that I wouldn't have come across otherwise - vaporwave, shoegaze, chiptune. Spotify playlists haven't come close. YouTube video 'playlists' - long videos that have curated songs - come close (playlists aren't as visible when you are searching or browsing) and if you listen to a few, you'll get similar in recommendations at least which are decent enough.
Funny thing, now that mixcloud is limiting users to ten shows total without paying them a subscription, 8tracks is once again the best option for cratediggers. That's where I'm sharing my stuff...
Funny thing, now that mixcloud is limiting users to ten shows total without paying them a subscription, 8tracks is once again the best option for cratediggers. That's where I'm sharing my stuff from now on. I'll probably drop the mixcloud page entirely.
I hadn't heard that about Mixcloud, that's a shame. I am really hoping for an 8tracks revival, it had almost shut down but got some money again and I'm just glad that it still exists in some form...
I hadn't heard that about Mixcloud, that's a shame. I am really hoping for an 8tracks revival, it had almost shut down but got some money again and I'm just glad that it still exists in some form still.
I discovered so many cool artists when the last.fm neighbors algorithm used to be better. You used to be able to stream neighbors radio and find new artists, too. The old algorithm got better at...
I discovered so many cool artists when the last.fm neighbors algorithm used to be better. You used to be able to stream neighbors radio and find new artists, too. The old algorithm got better at recommending music, the more you listened and rated songs. Sometime in the 20-teens, they changed the neighbors algorithm and it worked more like the algorithms we are used to now, based on your top listens and then genre tags that are just immediately related to what you've listened to. The old version seemed to do it per track, so for example, if you hated 95% of an artists catalog, but loved one of their songs, it would only match against that one song, and I assume form some sort of thumbprint or ratio that matched you to others on a per track basis.
Pandora actually has what I find to be the best music recommendation algorithm. Or at least they used to before I stopped using it. I think it’s because it was based more on matching music based...
Pandora actually has what I find to be the best music recommendation algorithm. Or at least they used to before I stopped using it. I think it’s because it was based more on matching music based on categorization on genre, key tonality, and a bunch of other such metadata associated to the songs rather than just machine learning on how well it’s associated with things others are listening to.
I found this to be a fairly interesting breakdown on why so much music starts to sounds same-y from a musician. He goes into how 'views' and 'listens' are counted among other topics....
I found this to be a fairly interesting breakdown on why so much music starts to sounds same-y from a musician. He goes into how 'views' and 'listens' are counted among other topics. Interestingly, he also point out the effects of algorithms of streaming services tend to lead to more and more homogenous suggestions and playlists.
This also depends on the musician and scale, but the style of modern pop music also lends itself to following formulas. When a song is written by one guy at a computer and shipped off to a pop...
This also depends on the musician and scale, but the style of modern pop music also lends itself to following formulas. When a song is written by one guy at a computer and shipped off to a pop singer to do the vocals on, it simplifies a lot of the traditional artistic process.
Before the rise of the DAW and VSTs, you had a house band and studio musicians. The songwriter could have given them a complete score or just chords, a lead sheet, and a lot of freedom. Regardless though, they rehearsed the song a whole lot before they hit record, they talked about it as a group, they jammed for fun between takes, and they came up with new parts while in the studio
An example that you may or may not consider pop, but is a song whose studio sessions I am familiar with, is “Like a Rolling Stone” by Dylan. Dylan originally wrote it in 3/4 but over the course of a few dozen takes, it evolved to the 4/4 time version we all know. Al Kooper, a young session guitarist at the time, was there as a friend of the producer and not supposed to play. During a break, he said “hey I got an idea for an organ part,” sat down, and played the iconic intro riff as an experiment. Dylan liked it so much he asked the organ to be turned up on the track and it’s arguably the most important non-vocal instrument in the song.
Musicians get bored doing the same thing over and over, they get bored, they innovate. When one guy’s computer does all the playing, that element of collaboration and evolution gets compressed.
Pop artists are still artists and they have their own processes, but this bit of a rant has been the result of me thinking about how a small number of producers and their computers dominate the song creation process for big studios, and how that’s qualitatively different from the analog era.
I think human curation has gone massively undervalued in almost every entertainment or entertainment-adjacent industry. Algorithms are cheap, and don't require any human labor, so of course...
I think human curation has gone massively undervalued in almost every entertainment or entertainment-adjacent industry. Algorithms are cheap, and don't require any human labor, so of course they're everywhere. The issue is simply that they suck at accurately modeling actual human listening tastes. They can't articulate why something might be interesting to invest your time in the way a human can; they just have arbitrary data points.
I appreciate that human curation still exists in many spaces, though. In most entertainment fields, there's still a healthy criticism/review culture, so I can always turn to those to find something I like. Even when it's not a large publication's main reviewer, it's nice to follow someone on YouTube, or follow a curator on Steam, for example. Getting an idea of why a living, breathing human thinks a piece of art is worthwhile is infinitely more valuable for my own discovery than an algorithm that judges based on sameness, so I'm glad it still exists in most places.
Pandora allows you to select a genre and it will algorithmically select artist and tracks within the genre. Unlike 10 years ago where everything eventually led to The Beatles, the algorithm works...
Pandora allows you to select a genre and it will algorithmically select artist and tracks within the genre. Unlike 10 years ago where everything eventually led to The Beatles, the algorithm works well (at least within trip hop, doom metal, stoner rock, psych and opera...)
Human curated content is making a comeback. I can heartily recommend NTS Radio (London) for its airplay and its infinite mixtapes. The Lot (NY), Psyched! (SF) and KEXP (Seattle) are also worthwhile.
So, it seems like it is going against the flow of the majority of the comments here, but, I think this is a pretty natural progression of art. When we have a new way to distribute art the art is...
So, it seems like it is going against the flow of the majority of the comments here, but, I think this is a pretty natural progression of art. When we have a new way to distribute art the art is going to eventually morph itself to fit inside of that form. Ex. Paintings and literature being adapted to codec format, song length shortening to under 4 minutes to fit on a record, the list goes on.
I think the interesting conversation around all of this is the involvement of corporate manipulation around trends more so than the art becoming homogenous after 2 decades of proliferate streaming. Should companies be allowed to use personal data to create these algorithms? Because they are useful (autocomplete text, location based searches, etc.) But at what cost to corporate control is it worth it? Idk. We already saw fb kind of shit the bed on having their algo TOO dialed in .
The reason algorithms force this to happen is the same reason the algorithms force images to outpace other content - upvote velocity, or in this case, listening velocity. Shorter tracks rack up more listens which immediately bias them in the data towards more plays. It's a simple math problem if Spotify cared to solve it. It is absolutely not what the consumers want. The consumers aren't even paying much attention. They just loop their faves or bounce on to a playlist curated by humans, not algorithms.
Algorithms are useless in this area, really. No algorithm can ever understand music because it has no emotional context to appreciate that music from in the first place. Human recommendations are better than algorithms when it comes to music. Spotify is counting numbers when they should be building a social network for their listeners to share music instead.
Legit - it's a big problem for me with Spotify because a lot of the time I'll like or add to a playlist a song from a band, while not actually liking their other stuff that much... but then I get them shoved into any "random" playlist...
I seem to get stuck in "genre holes" on Spotify. Sometimes my recommendations kick ass and I find new favorites. Most of the time, however, I feel like I'm trapped by one sound and Spotify won't let me break out. My Discover Weekly seems designed to push me back to things I already like rather than introducing me to anything new.
As I type this, Jungle by Jimi Hendrix is number three on my Discover Weekly. Jungle (this version anyway) is an instrumental in the middle of Both Sides of the Sky, it's not a song I'd listen to on its own like, say, Red House. Furthermore, I've liked 17 Hendrix songs; pretty good chance I've heard Jungle once or twice. Why am I being recommended Hendrix anyway? I don't need Spotify to introduce me to Jimi Hendrix, I need Spotify to show me new stuff that I don't already know about.
I find what works best is to make a playlist of at least 20 tracks of stuff you like, then make a copy of that playlist, hit the copy with the 'enhance' feature, and check out the songs it auto-populated. This is kinda forcing it to make you a discovery weekly on demand, as if that playlist were your liked songs.
While the songs themselves may not be the best choices, the artists it throws at you are probably worth a quick brush of their top tracks. You can keep on repeating this process a couple of times, and it will eventually run out of fresh content to enhance with and start repeating the same artists. That's the time to create a new playlists of other tracks you like and try it again.
Another way is to track down other music hounds on Spotify that share your tastes and follow them. The genre playlists being auto-generated by the service itself aren't useless, but they aren't deep cuts either. Other user playlists don't have that problem. Where else but spotify can you find 2,177 songs that 'french people would play at a party or just with friends around' still collecting new tracks all the time.
It annoys me when artists disappear from their service. I really don't care that Burger Records had some serious issues with exploitation - they paid for it and it's none of my business. I do care that their entire catalog of garage rock and other oddities, encompassing dozens of artists, has vanished from the service as if every single artist was personally involved in the scandal that took their publisher down. :/
I think Last.fm sort of hit on one of the most interesting ideas for this: let users tag music. Then someone could choose to view/listen to a tag, and find new recommendations that many people chose to tag with a certain keyword. It works well for niche genres, which conventional music metadata doesn't really specify. "Dance Music" isn't helpful when I'm looking for Tech House or 90s Eurodance.
The peak for playlists for me was early 2010s 8tracks, which worked because of tags like you mentioned. I discovered all sorts of genres on there that I wouldn't have come across otherwise - vaporwave, shoegaze, chiptune. Spotify playlists haven't come close. YouTube video 'playlists' - long videos that have curated songs - come close (playlists aren't as visible when you are searching or browsing) and if you listen to a few, you'll get similar in recommendations at least which are decent enough.
Funny thing, now that mixcloud is limiting users to ten shows total without paying them a subscription, 8tracks is once again the best option for cratediggers. That's where I'm sharing my stuff from now on. I'll probably drop the mixcloud page entirely.
I hadn't heard that about Mixcloud, that's a shame. I am really hoping for an 8tracks revival, it had almost shut down but got some money again and I'm just glad that it still exists in some form still.
I discovered so many cool artists when the last.fm neighbors algorithm used to be better. You used to be able to stream neighbors radio and find new artists, too. The old algorithm got better at recommending music, the more you listened and rated songs. Sometime in the 20-teens, they changed the neighbors algorithm and it worked more like the algorithms we are used to now, based on your top listens and then genre tags that are just immediately related to what you've listened to. The old version seemed to do it per track, so for example, if you hated 95% of an artists catalog, but loved one of their songs, it would only match against that one song, and I assume form some sort of thumbprint or ratio that matched you to others on a per track basis.
Pandora actually has what I find to be the best music recommendation algorithm. Or at least they used to before I stopped using it. I think it’s because it was based more on matching music based on categorization on genre, key tonality, and a bunch of other such metadata associated to the songs rather than just machine learning on how well it’s associated with things others are listening to.
I found this to be a fairly interesting breakdown on why so much music starts to sounds same-y from a musician. He goes into how 'views' and 'listens' are counted among other topics. Interestingly, he also point out the effects of algorithms of streaming services tend to lead to more and more homogenous suggestions and playlists.
This also depends on the musician and scale, but the style of modern pop music also lends itself to following formulas. When a song is written by one guy at a computer and shipped off to a pop singer to do the vocals on, it simplifies a lot of the traditional artistic process.
Before the rise of the DAW and VSTs, you had a house band and studio musicians. The songwriter could have given them a complete score or just chords, a lead sheet, and a lot of freedom. Regardless though, they rehearsed the song a whole lot before they hit record, they talked about it as a group, they jammed for fun between takes, and they came up with new parts while in the studio
An example that you may or may not consider pop, but is a song whose studio sessions I am familiar with, is “Like a Rolling Stone” by Dylan. Dylan originally wrote it in 3/4 but over the course of a few dozen takes, it evolved to the 4/4 time version we all know. Al Kooper, a young session guitarist at the time, was there as a friend of the producer and not supposed to play. During a break, he said “hey I got an idea for an organ part,” sat down, and played the iconic intro riff as an experiment. Dylan liked it so much he asked the organ to be turned up on the track and it’s arguably the most important non-vocal instrument in the song.
Musicians get bored doing the same thing over and over, they get bored, they innovate. When one guy’s computer does all the playing, that element of collaboration and evolution gets compressed.
Pop artists are still artists and they have their own processes, but this bit of a rant has been the result of me thinking about how a small number of producers and their computers dominate the song creation process for big studios, and how that’s qualitatively different from the analog era.
I think human curation has gone massively undervalued in almost every entertainment or entertainment-adjacent industry. Algorithms are cheap, and don't require any human labor, so of course they're everywhere. The issue is simply that they suck at accurately modeling actual human listening tastes. They can't articulate why something might be interesting to invest your time in the way a human can; they just have arbitrary data points.
I appreciate that human curation still exists in many spaces, though. In most entertainment fields, there's still a healthy criticism/review culture, so I can always turn to those to find something I like. Even when it's not a large publication's main reviewer, it's nice to follow someone on YouTube, or follow a curator on Steam, for example. Getting an idea of why a living, breathing human thinks a piece of art is worthwhile is infinitely more valuable for my own discovery than an algorithm that judges based on sameness, so I'm glad it still exists in most places.
Pandora allows you to select a genre and it will algorithmically select artist and tracks within the genre. Unlike 10 years ago where everything eventually led to The Beatles, the algorithm works well (at least within trip hop, doom metal, stoner rock, psych and opera...)
Human curated content is making a comeback. I can heartily recommend NTS Radio (London) for its airplay and its infinite mixtapes. The Lot (NY), Psyched! (SF) and KEXP (Seattle) are also worthwhile.
I do love KEXP; consequently I greatly appreciate the NTS Radio, The Lot, and Psyched! suggestions.
So, it seems like it is going against the flow of the majority of the comments here, but, I think this is a pretty natural progression of art. When we have a new way to distribute art the art is going to eventually morph itself to fit inside of that form. Ex. Paintings and literature being adapted to codec format, song length shortening to under 4 minutes to fit on a record, the list goes on.
I think the interesting conversation around all of this is the involvement of corporate manipulation around trends more so than the art becoming homogenous after 2 decades of proliferate streaming. Should companies be allowed to use personal data to create these algorithms? Because they are useful (autocomplete text, location based searches, etc.) But at what cost to corporate control is it worth it? Idk. We already saw fb kind of shit the bed on having their algo TOO dialed in .