Oh my God! I've been wondering about this exact thing for years! In fact, I thought I might had made a significant contribution to the songs Spotify rank. I've even talked about it on the pavement...
Oh my God! I've been wondering about this exact thing for years! In fact, I thought I might had made a significant contribution to the songs Spotify rank. I've even talked about it on the pavement subreddit a few times, which I'll copy below. You have no idea how excited I was to see this article. Thanks for submitting.
Here's a comment I made about the song a few month back:
Holy shit, I've been waiting for an opportunity to tell this story for years. I know it's going to sound insane, but I think I might've played a huge role in the song's rise. I know, there's probably no way, but hear me out. I first got really into Pavement back in early 2016. At the time, the top songs on Spotify had a couple million plays. I think Range Life was somewhere around 4 or 5 million, Cut Your Hair and Goldsoundz were around 2 or 3 million, then the rest of the top-10 all hovered around a million, give or take a few hundred thousand. After a few months of focusing on Quarantine the Past and acquiring a taste for the laid-back style, I started to work my way through the discography and saved any song that I enjoyed along the way. That's when I found Harness Your Hopes and I. FUCKING. LOVED IT. I'm not kidding or embellishing when I say I played it at least 20 times a day for about a month or two. 20 might even be an understatement, as insane as it sounds. I just loved the simple rhythm and the way the lyrics all seemed to fall into each other - it reminded me a lot of my favorite Sublime song, New Realization. Anyway, in addition to actively listening to the song hundreds (possibly over a thousand) of times in a six-week span, I also fell asleep to a short playlist I made that featured it, and it would play on repeat overnight (this happened maybe three or four times in total). After a few months, as my obsession spread to basically every Pavement song, I finally noticed one day that Harness Your Hopes had cracked the top ten. The crazy thing is it had significantly fewer plays than every other song. Like close to half of the next highest one, but it was at seven. I know its a stretch to suggest that I singlehandedly put it in the top-10, but I do wonder if I gave it a boost to end up on a mainstream playlist where non-pavement fans would give it plays. I would be shocked if there were people listening to the band as often as I did at that time, let alone that song. I wasn't working very often, but when I did work, I was alone in a service bar blasting Pavement for my entire shift. And as far as my personal life goes, I was going through some shit (turns out music isn't the only thing I get addicted to) so I would go on really long walks then play video games while listening to Pavement the whole time. I've always wanted to ask someone at Spotify if something like that could really affect the algorithm.
I remember really wanting to post about it when it cracked the top-10 (I hadn't found this sub yet) and I was always waiting for a relevant moment to tell this story. Anyway, thanks for finally giving me a chance to talk about how obsessed I was with this song.
I can personally attest to how much Spotify loves this song. I'm not the biggest Pavement fan, though Gold Soundz is on a few of my playlists - but I do listen to a lot of music that may be...
I can personally attest to how much Spotify loves this song. I'm not the biggest Pavement fan, though Gold Soundz is on a few of my playlists - but I do listen to a lot of music that may be considered Pavement-adjacent - and sure enough, Harness Your Hopes seems to get into nearly every one of my daily mixes. It's not a bad track but I do tend to use a skip up on it now because otherwise I'd hear it every day.
I'm a Pavement fan and I haven't been reco'd Harness Your Hopes, but I have definitely had "Strange" by Galaxie 500 (the other song in the article) come up in several personalized playlists. I've...
I'm a Pavement fan and I haven't been reco'd Harness Your Hopes, but I have definitely had "Strange" by Galaxie 500 (the other song in the article) come up in several personalized playlists. I've also heard it playing in a cafe once or twice. I always just assumed it was their one hit or something, but it wasn't even a single.
That being said, I work on product recommendation (for e-commerce, not music), so I found this article pretty fascinating. The theory that the recommender happened to surface genuinely good songs that had previously fallen under the radar is compelling but almost certainly not the whole story. I also suspect @AnthonyB's comment about streaming Harness Your Hopes an inordinate amount of time did not contribute much to it either (in theory, recommender systems should not be so easy for a single user to influence, and Spotify's got a pretty big user base listening to a lot of songs).
I also am suspicious about the idea that these songs are being recommended because they sound very similar to lots of songs. Spotify has lots of different personalized features (custom playlists, radio/autoplay, "Browse" screen) and I don't know what types of algorithms they use for each one, but I suspect many are collaborative-filtering based. When CF is biased, it tends to be biased towards the most popular items, not the most "average" ones.
That being said, I don't have a great explanation for why these songs would be so heavily recommended. It could be a bug, or an unexpected edge case, or it could be that the system really does find hidden gems.
(I also think people get a bit too wrapped up in an artist's top songs. Right now, for instance, the #2 Arthur Russell song has 537k plays, while #1 has 9.5 million and #3 has 3.1 million, so clearly they're not strictly ranked for historical popularity. I suspect they rank by most listens recently, and probably deduplicate if multiple copies of the same track end up in the top.)
Thanks for your insight. I'm actually currently studying a module in Recommender Systems at University and it's always interesting to see ideas fall out of the PowerPoints and collide with the...
Thanks for your insight. I'm actually currently studying a module in Recommender Systems at University and it's always interesting to see ideas fall out of the PowerPoints and collide with the real world.
It's a fascinating topic that's evolving very quickly. I'm also taking a class in recommender systems as part of my masters, in addition to my work, so I've been constantly going back to my...
It's a fascinating topic that's evolving very quickly. I'm also taking a class in recommender systems as part of my masters, in addition to my work, so I've been constantly going back to my coworkers with a new idea from the latest lecture or reading.
A similar thing happened on youtube with Ryo Fukui's album Scenery. It's a Japanese jazz album, pretty middle of the road piano trio and not at all well known outside of Japan. As jazz goes it's...
A similar thing happened on youtube with Ryo Fukui's album Scenery. It's a Japanese jazz album, pretty middle of the road piano trio and not at all well known outside of Japan. As jazz goes it's fairly mediocre IMO. (This is a point of never ending debate in /r/jazz.) But a few years ago, seemingly out of nowhere, it started popping up in the recommendations sidebar on nearly every jazz video on youtube. Now the full album has over ten million views, has been reissued on vinyl several times after many years out of print, and is now considered by many to be a rediscovered classic. Go figure.
Oh my God! I've been wondering about this exact thing for years! In fact, I thought I might had made a significant contribution to the songs Spotify rank. I've even talked about it on the pavement subreddit a few times, which I'll copy below. You have no idea how excited I was to see this article. Thanks for submitting.
Here's a comment I made about the song a few month back:
Holy shit, I've been waiting for an opportunity to tell this story for years. I know it's going to sound insane, but I think I might've played a huge role in the song's rise. I know, there's probably no way, but hear me out. I first got really into Pavement back in early 2016. At the time, the top songs on Spotify had a couple million plays. I think Range Life was somewhere around 4 or 5 million, Cut Your Hair and Goldsoundz were around 2 or 3 million, then the rest of the top-10 all hovered around a million, give or take a few hundred thousand. After a few months of focusing on Quarantine the Past and acquiring a taste for the laid-back style, I started to work my way through the discography and saved any song that I enjoyed along the way. That's when I found Harness Your Hopes and I. FUCKING. LOVED IT. I'm not kidding or embellishing when I say I played it at least 20 times a day for about a month or two. 20 might even be an understatement, as insane as it sounds. I just loved the simple rhythm and the way the lyrics all seemed to fall into each other - it reminded me a lot of my favorite Sublime song, New Realization. Anyway, in addition to actively listening to the song hundreds (possibly over a thousand) of times in a six-week span, I also fell asleep to a short playlist I made that featured it, and it would play on repeat overnight (this happened maybe three or four times in total). After a few months, as my obsession spread to basically every Pavement song, I finally noticed one day that Harness Your Hopes had cracked the top ten. The crazy thing is it had significantly fewer plays than every other song. Like close to half of the next highest one, but it was at seven. I know its a stretch to suggest that I singlehandedly put it in the top-10, but I do wonder if I gave it a boost to end up on a mainstream playlist where non-pavement fans would give it plays. I would be shocked if there were people listening to the band as often as I did at that time, let alone that song. I wasn't working very often, but when I did work, I was alone in a service bar blasting Pavement for my entire shift. And as far as my personal life goes, I was going through some shit (turns out music isn't the only thing I get addicted to) so I would go on really long walks then play video games while listening to Pavement the whole time. I've always wanted to ask someone at Spotify if something like that could really affect the algorithm.
I remember really wanting to post about it when it cracked the top-10 (I hadn't found this sub yet) and I was always waiting for a relevant moment to tell this story. Anyway, thanks for finally giving me a chance to talk about how obsessed I was with this song.
I can personally attest to how much Spotify loves this song. I'm not the biggest Pavement fan, though Gold Soundz is on a few of my playlists - but I do listen to a lot of music that may be considered Pavement-adjacent - and sure enough, Harness Your Hopes seems to get into nearly every one of my daily mixes. It's not a bad track but I do tend to use a skip up on it now because otherwise I'd hear it every day.
I'm a Pavement fan and I haven't been reco'd Harness Your Hopes, but I have definitely had "Strange" by Galaxie 500 (the other song in the article) come up in several personalized playlists. I've also heard it playing in a cafe once or twice. I always just assumed it was their one hit or something, but it wasn't even a single.
That being said, I work on product recommendation (for e-commerce, not music), so I found this article pretty fascinating. The theory that the recommender happened to surface genuinely good songs that had previously fallen under the radar is compelling but almost certainly not the whole story. I also suspect @AnthonyB's comment about streaming Harness Your Hopes an inordinate amount of time did not contribute much to it either (in theory, recommender systems should not be so easy for a single user to influence, and Spotify's got a pretty big user base listening to a lot of songs).
I also am suspicious about the idea that these songs are being recommended because they sound very similar to lots of songs. Spotify has lots of different personalized features (custom playlists, radio/autoplay, "Browse" screen) and I don't know what types of algorithms they use for each one, but I suspect many are collaborative-filtering based. When CF is biased, it tends to be biased towards the most popular items, not the most "average" ones.
That being said, I don't have a great explanation for why these songs would be so heavily recommended. It could be a bug, or an unexpected edge case, or it could be that the system really does find hidden gems.
(I also think people get a bit too wrapped up in an artist's top songs. Right now, for instance, the #2 Arthur Russell song has 537k plays, while #1 has 9.5 million and #3 has 3.1 million, so clearly they're not strictly ranked for historical popularity. I suspect they rank by most listens recently, and probably deduplicate if multiple copies of the same track end up in the top.)
Thanks for your insight. I'm actually currently studying a module in Recommender Systems at University and it's always interesting to see ideas fall out of the PowerPoints and collide with the real world.
It's a fascinating topic that's evolving very quickly. I'm also taking a class in recommender systems as part of my masters, in addition to my work, so I've been constantly going back to my coworkers with a new idea from the latest lecture or reading.
A similar thing happened on youtube with Ryo Fukui's album Scenery. It's a Japanese jazz album, pretty middle of the road piano trio and not at all well known outside of Japan. As jazz goes it's fairly mediocre IMO. (This is a point of never ending debate in /r/jazz.) But a few years ago, seemingly out of nowhere, it started popping up in the recommendations sidebar on nearly every jazz video on youtube. Now the full album has over ten million views, has been reissued on vinyl several times after many years out of print, and is now considered by many to be a rediscovered classic. Go figure.
Really excellent contribution, thanks for posting :)