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

51–60 of 275 posts

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

#51

Post by Systemic33 »

a254613e wrote: post #6

> Unlike Netflix, though, Spotify doesn’t have those stars with which users rate their music. Instead, Spotify’s data is implicit feedback I wish they kept it simpler with ratings, and explicit instead of implicit. The discovery weekly playlists are absolutely horrible for me to the point that I don't even bother checking them nowadays. What works better for me for discovering new music with spotify is right clicking…

Explicit and implicit rating systems will never fit everyone.

But I will go as far and say that implicit rating fits a lot more than explicit, because it doesn't require the user to do extra work on top of the base goal of listing to "good" music.

I feel like explicit rating scales very poorly with catalogue size. So when you have music, and as much music as Spotify has, then the work effort of explicitly rating your taste becomes too big, you start to not bother, and quality of rating becomes poor.

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

#52

Post by emerongi »

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.

For some reason every week I get like 2-5 songs with the same beat, but a bit remixed or covered by someone else. It's trying way too hard to recommend similar songs sometimes.

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

#53

Post by marcus_holmes »

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 traffic.

Or have I got this wrong?

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

#54

Post by cr1895 »

confounded wrote: post #41
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’ve noticed that there are certain seemingly-esoteric songs that many people are assigned, and subsequently feel let down upon realizing that they’re (currently) common assignations. It wasn’t because the algorithms recognized their excellent personal taste. “ Didn’t I ” by Darondo seems to be a good example of this. A fantastic, forgotten soul song (and from the Bay Area!) that was assumed to be a recognition of my…

>“Didn’t I” by Darondo seems to be a good example of this. A fantastic, forgotten soul song (and from the Bay Area!) that was assumed to be a recognition of my friend’s unique taste in his car... until the other four people inside revealed they’d all had it in the last two weeks, too.

It could be popularized by DJs who are playing it, e.g. Four Tet, whose personal playlist on Spotify now has close to 39 thousand followers.

At least in that case specifically, I doubt it's due to distribution rights. There are no new releases indicated on Discogs, which seems likely if someone new acquired them.

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

#55

Post by tjoff »

Jdam wrote: post #30
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.

So the dreaded filter bubble also is an issue for Spotify ;)

I'm afraid it is so extreme that I feel lots of my discover weekly recommendations are based on solely of what I listened on discover weekly last week.

Which would be quite bad, but that is how it often feels.

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

#56

Post by cr1895 »

ricardobeat wrote: post #23

Discover Weekly works well enough for me - I do wish I could exclude devices from influencing that playlist as what I listen while gaming on PS4 is completely different from the rest of the day. It also seems to have a bias towards “big” commercial releases. Spotify’s “intelligence” in general is a huge let down though. Radio stations are extremely limited - more like 15-song static playlists indefinitely on repeat!…

>It also seems to have a bias towards “big” commercial releases.

I haven't noticed this bias at all. It might be due to our own music preferences.

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

#57

Post by Fabled »

In case you didn't know, Spotify categorizes music into genres behind the scenes. You can use this site to find out what genres your favorite artists are categorized as: http://everynoise.com/engenremap.html

You can use this information to then check out Spotify's auto-generated playlists for each genre. They have at least three types for each one: "The Sound of ", containing definitive representation of the genre, "The Pulse of ", containing songs that fans of the genre listen to now, and "The Edge of ", with unpopular songs (not necessarily of the same genre) that fans of the genre listen to.

This has been a great way for me to find new music that I like, especially "The Pulse". I even created a small script that takes a spotify playlist as input, parses all the artists, converts to genres, and creates a new playlist with "recommendations" based on the Pulse playlists, with each genre represented based on its percentage in the initial playlist.

Basically, I never use Discover Weekly, because I know it will eventually converge just like all my Pandora stations that cycle through the same 20 songs after a few months.

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

#58

Post by iddqd »

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 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 least give them some sense of direction.

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

#59

Post by KozmoNau7 »

iddqd wrote: post #58
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 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…

You should branch out from only listening to "Discover Weekly". Use the daily mixes, make your own playlists.

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

#60

Post by KozmoNau7 »

Fabled wrote: post #57

In case you didn't know, Spotify categorizes music into genres behind the scenes. You can use this site to find out what genres your favorite artists are categorized as: http://everynoise.com/engenremap.html You can use this information to then check out Spotify's auto-generated playlists for each genre. They have at least three types for each one: "The Sound of ", containing definitive representation of the genre, "…

I only use the Discover Weekly list once in a while. I prefer using the Daily Mixes, and my own playlists. It seems to make Spotify match my taste quite well.
Post reply on HN