Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Spotify’s Discover Weekly: How machine learning finds new music

hackernoon.com

Post reply on HN
Print view View original

121–130 of 275 posts

Re: Spotify’s Discover Weekly: How machine learning finds new music

#121

Post by mixedbit »

arkadiytehgraet wrote: post #80

If there is at least one song that I like in Discover Weekly, then I consider it to be a good mix. However, this happens once a month at best. After reading the article and all three methods for recommendations, I cannot understand how on Earth it comes up with its suggestions. Do I feed it bad data? Does it not take into account the relative "weight" of a particular track, i.e. my preference to listen to it multiple…

My friend likes and uses the Discover Weekly and to get good recommendations he uses different Spotify account when he plays something for his kids or plays some generic background music at home.

Re: Spotify’s Discover Weekly: How machine learning finds new music

#122

Post by graphitezepp »

dx034 wrote: post #106
iddqd wrote: post #58

Earlier quoted context omitted.

I used to listen exclusively to Discover Weekly for a while. One week a finish rap song made it to the list, and I skipped past it every time it came on. Next week there was 2 finish rap songs. Then 5. Eventually half of my discovery list was finish rap, something I have no interest of. Canceled my subscription shortly after. Their algorithms are feeding themselves. I wish they had a dislike button so I could at leas…

I have the exact same thing with Dutch rap music. I'm not Dutch, don't speak Dutch and have never listened to Dutch music. But Spotify consistently puts 2-4 Dutch songs on my Discover Weekly.

This is hilarious to me. On the other side, I have been getting super into French rap music, and Spotify refuses to notice.

Re: Spotify’s Discover Weekly: How machine learning finds new music

#123

Post by graphitezepp »

chrissnell wrote: post #71
Asdfbla wrote: post #29

Sometimes I wish Spotify added a bit more 'noise' to their recommendations, so to speak. If I don't listen to much music except Discover Weekly for a few weeks, I (subjectively) find that what's recommended to me more or less sounds the same after a while. Either they are afraid to insert new things that stray too far from an optimal recommendation or they forget too much of my listening history.

I think this might be a huge problem for them and I’m pretty sure that they’re aware of it. The lack of variety seems to span all genres and radio stations. I’m really into Texas Country, Americana, and Red Dirt music. These are distinct sub-genres under “alt-Country” music but from the Spotify perspective, they’re all basically the same 40-50 songs. I can’t tell the difference between the Reckless Kelly, Robert Earl…

I have had significantly worse luck with Pandora even, but haven't used it in ages. Rdio, before it went bankrupt, worked the best for me.

Re: Spotify’s Discover Weekly: How machine learning finds new music

#124

Post by KozmoNau7 »

mixedbit wrote: post #121
arkadiytehgraet wrote: post #80

If there is at least one song that I like in Discover Weekly, then I consider it to be a good mix. However, this happens once a month at best. After reading the article and all three methods for recommendations, I cannot understand how on Earth it comes up with its suggestions. Do I feed it bad data? Does it not take into account the relative "weight" of a particular track, i.e. my preference to listen to it multiple…

My friend likes and uses the Discover Weekly and to get good recommendations he uses different Spotify account when he plays something for his kids or plays some generic background music at home.

He can use the "private session" option for that instead.

Re: Spotify’s Discover Weekly: How machine learning finds new music

#125

Post by rodorgas »

mattmanser wrote: post #87
arkadiytehgraet wrote: post #80

If there is at least one song that I like in Discover Weekly, then I consider it to be a good mix. However, this happens once a month at best. After reading the article and all three methods for recommendations, I cannot understand how on Earth it comes up with its suggestions. Do I feed it bad data? Does it not take into account the relative "weight" of a particular track, i.e. my preference to listen to it multiple…

I don't know, I have friends who listen to some music I like, but lots of music I don't. If it's trying to recommend by that, it's going to miss a lot. And even within my own listening, there's plenty of bands where I like one song, but hate the rest of the album. There's bands where I'm massive fans of certain albums, but only like one song on others. Music's a weird one. And taste changes. Last night I was listenin…

That's why I'm skeptical in machines recommending music. Deezer focuses more in Editors curated playlists (lots of them) and users playlists recommendation (instead of single songs). I've switched to Spotify last year because UX is better, but I feel I discovered less music I actually enjoy.

Re: Spotify’s Discover Weekly: How machine learning finds new music

#126

Post by graphitezepp »

smnscu wrote: post #48
thirdsun wrote: post #47

Earlier quoted context omitted.

Rdio had an option to address that: Its radio / "play similar stuff"-feature would let you select how far you wanted to stray from the original source of the generated playlist. If I remember correctly it went from "same artist" to "adventurous" in 5 steps.

Would you please stop reminding me how much better Rdio was :(

Nobody understands when I complain about losing Rdio (which is often), it was sooooo much better and nobody knew/cared.

Re: Spotify’s Discover Weekly: How machine learning finds new music

#127

Post by KozmoNau7 »

chrissnell wrote: post #71
Asdfbla wrote: post #29

Sometimes I wish Spotify added a bit more 'noise' to their recommendations, so to speak. If I don't listen to much music except Discover Weekly for a few weeks, I (subjectively) find that what's recommended to me more or less sounds the same after a while. Either they are afraid to insert new things that stray too far from an optimal recommendation or they forget too much of my listening history.

I think this might be a huge problem for them and I’m pretty sure that they’re aware of it. The lack of variety seems to span all genres and radio stations. I’m really into Texas Country, Americana, and Red Dirt music. These are distinct sub-genres under “alt-Country” music but from the Spotify perspective, they’re all basically the same 40-50 songs. I can’t tell the difference between the Reckless Kelly, Robert Earl…

Go discover some new artists by yourself, use the related artists function, branch out.

I'm pretty sure that'll shake up the recommendations a bit.

Re: Spotify’s Discover Weekly: How machine learning finds new music

#128

Post by whiskeySix »

chrissnell wrote: post #71
Asdfbla wrote: post #29

Sometimes I wish Spotify added a bit more 'noise' to their recommendations, so to speak. If I don't listen to much music except Discover Weekly for a few weeks, I (subjectively) find that what's recommended to me more or less sounds the same after a while. Either they are afraid to insert new things that stray too far from an optimal recommendation or they forget too much of my listening history.

I think this might be a huge problem for them and I’m pretty sure that they’re aware of it. The lack of variety seems to span all genres and radio stations. I’m really into Texas Country, Americana, and Red Dirt music. These are distinct sub-genres under “alt-Country” music but from the Spotify perspective, they’re all basically the same 40-50 songs. I can’t tell the difference between the Reckless Kelly, Robert Earl…

Pandora? Really? I've found Spotify to be orders of magnitude better than Pandora for both variety and discovery. Pandora was the first to have competent discovery functionality, and it was impressive for its time, but I'm wondering if this is "good old days" syndrome speaking.

Re: Spotify’s Discover Weekly: How machine learning finds new music

#129

Post by dontreact »

marcus_holmes wrote: post #53
dontreact wrote: post #2

Pretty cool to see them go after the cold start problem with audio models. Often measuring the impact of having cold start vs. not is very difficult because it only affects the long tail of tracks at first, but eventually it could change the dynamics of discovery in the system.

It occurred to me, though, that this is wide open to hackery. Find a popular song that is likely to be on a lot of people's playlists. Make a new song that is a close match to it for raw audio modelling. Launch on Spotify. Yeah you'll probably only get one or two plays per user, but spread over a large number of users (who all listen to the whole Discover playlist every Monday, like I do) it's still significant traff…

I imagine that once you get to any significant number of plays (N=1000, which would not be a significant payout on spotify at all), the collaborative filter model would take over and if none of the people who listened to it caused further positive signals as mentioned (visiting the artist page, repeat listens) then the track would stop being recommended pretty quickly.

Also, what you are describing sounds like what producers already do because people actually like it: find the latest trends in sound and copy them, (but hopefully with a fresh twist so that people get into it) :)

Re: Spotify’s Discover Weekly: How machine learning finds new music

#130

Post by dom96 »

dom96 wrote: post #83
arkadiytehgraet wrote: post #80

If there is at least one song that I like in Discover Weekly, then I consider it to be a good mix. However, this happens once a month at best. After reading the article and all three methods for recommendations, I cannot understand how on Earth it comes up with its suggestions. Do I feed it bad data? Does it not take into account the relative "weight" of a particular track, i.e. my preference to listen to it multiple…

I have the same experience. Although certainly it seems that my mood affects the kind of music that I enjoy. There were times when I listened to my discover weekly playlist on Monday, didn't like any of the songs, then listened again 6 days later and liked the vast majority. This is difficult to measure of course, and is purely anecdotal. I am also wondering whether there is simply no more music out there that I enjo…

Something to add to this: it's also incredibly common for Discover Weekly to suggest songs which I have already listened to hundreds of times, does anyone else have this experience? I don't understand why this would be unless their algorithm simply cannot find anything else. I'd rather they just be honest instead of selecting songs I already know.
Post reply on HN