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

171–180 of 275 posts

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

#171

Post by mediocrejoker »

mulletbum wrote: post #151
luckydata wrote: post #142

I spent quite a bit of time thinking about the recommendation algorithm and (after falling into a death metal hole I can't seem to escape on Spotify) I came to the conclusion that their analysis of the audio content of the song is way too shallow. You listen to songs because they have a common harmonic structure, or rythm, but not necessarily the same spectrum. That's why a metal fan might dig a cover of Metallica by…

I just always assumed it would be better to match up peoples likes/dislikes against one another. For example, I might really dig Brittany Spears, but I also like Pantera. Considering how those types of music have nothing to do with each other, it would be much better to match me up against other people who have similar taste in music as me. So let's say for example there are 1,000 people who like Spears and Pantera (…

I am not a data scientist, but this seems like a very obvious solution to me as well so I wonder if there is a reason it's not done this way.

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

#172

Post by nobleach »

I find that my music tastes can vary too much for Spotify. While working, I mostly listen to Electronic music. But I can also take a hard right turn and listen to Hall and Oates... or Fleetwood Mac, or Steely Dan. Then some form of 80's throwback. But as far as recommendations go, I really only want to expand in that first genre of Electronic Music. I don't want Jethro Tull to be recommended to me. My feed used to be full of really cool stuff, but I find that now I have several lines of "Because you listened to one song by The Pointer Sisters...."

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

#173

Post by acdha »

iaml wrote: post #150
acdha wrote: post #148

I find the premise fascinating because of how bad the Discover Weekly recommendations were when I was looking for a replacement for Rdio. Rdio did a great job of suggesting new music but with Spotify I found myself constantly skipping tracks – combined with not having a way to play album tracks in order[1] I found no justification for using, much less paying for the service. 1. Yes, they claim to offer that for paid…

It's really inconsistent - sometimes I add pretty much every single track from the discover weekly, and sometimes it's just a stream of garbage for multiple weeks straight. It also needs some time to adjust for your tastes - I got decent compilations only like 1 month in.

I bailed after 3 months but the bimodal response distribution here is making me wonder whether they did have something like a broken A/B test, monitoring metrics which suppresses “outliers”, etc. which is keeping the team from seeing the subset of users who aren’t getting good results.

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

#174

Post by micahjonas »

markussss wrote: post #140

The biggest problem with the Discover Weekly is its inability to understand _why_ you're listening to a specific subset of music. It might not be that your taste in music suddenly changed, or that you discovered a new genre that you're incredibly interested in, even though your most listened to genres or songs changed for a few weeks, or your listening patterns changed for a few weeks. A few examples: I'm Norwegian,…

Maybe don't listen to the discovery weekly playlist for the week and just listen to other playlists you like. If you listen to the genre because it's in your weekly playlist, than spotify is probably going to think you like those songs.

Search for some open playlist you like and don't listen to the discovery weekly. In my experience spotify is fairly good a taking the hint.

One alternative would be the daily selection which seems to be more sorted for genres.

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

#175

Post by pj_mukh »

markussss wrote: post #140

The biggest problem with the Discover Weekly is its inability to understand _why_ you're listening to a specific subset of music. It might not be that your taste in music suddenly changed, or that you discovered a new genre that you're incredibly interested in, even though your most listened to genres or songs changed for a few weeks, or your listening patterns changed for a few weeks. A few examples: I'm Norwegian,…

I really wish I could get Spotify to "forget" or "reset" my preferences. When I first started using Spotify, I listened to a lot of slow calming music (because I was working). Now ALL my Discover weekly's are just 'Iron and Wine', 'The National' derivatives. Its made Discover Weekly useless to me.

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

#176

Post by ghaff »

mulletbum wrote: post #151
luckydata wrote: post #142

I spent quite a bit of time thinking about the recommendation algorithm and (after falling into a death metal hole I can't seem to escape on Spotify) I came to the conclusion that their analysis of the audio content of the song is way too shallow. You listen to songs because they have a common harmonic structure, or rythm, but not necessarily the same spectrum. That's why a metal fan might dig a cover of Metallica by…

I just always assumed it would be better to match up peoples likes/dislikes against one another. For example, I might really dig Brittany Spears, but I also like Pantera. Considering how those types of music have nothing to do with each other, it would be much better to match me up against other people who have similar taste in music as me. So let's say for example there are 1,000 people who like Spears and Pantera (…

One problem you run into is that popular artists/titles don't really give a lot of insight into what else you'll like. This applies to a lot of recommendation areas. You like Star Wars. Great. You and a billion other people. And, by the way, I can also recommend Star Wars to you just by looking at the box office returns.

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

#177

Post by bubblethink »

Off topic: Does anyone remember thesixtyone ? I loved that site, and a lot of the music that I listen to was discovered there. It has a very "the day the music died" sort of feeling for me, when they shut. They were healthy until 2010. Then it shut for while, was on life support for a while, and was officially killed sometime last year. It was very simple and light, and had interesting music. I never found (or put too much effort honestly) to find a good replacement among the next wave of music websites. I can't stand the whole log in with facebook and enable drm in chrome to listen to this song thing. Just want some new interesting, indie music. Suggestions welcome.

Further off topic: Among the commercial providers, I found this Indian music site very refreshing (https://gaana.com/). No sign-in, no flash, no widevine.

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

#178

Post by georgiecasey »

steve-benjamins wrote: post #89

As an artist, Discover Weekly has been the best thing to ever happen to me. Every Monday I get a big infusion of listeners (around 5,000)— many of who stick around and check out my other music :) Prior to that, the best press I could get was the tedious process of cold-emailing bloggers (a practice which is now dying off).

ya i think blogs are dying. i used to visit hypemachine daily from 2007-2014 but discover weekly has taken over that music discovery job.

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

#179

Post by eeeeeeeeeeeee »

nobleach wrote: post #172

I find that my music tastes can vary too much for Spotify. While working, I mostly listen to Electronic music. But I can also take a hard right turn and listen to Hall and Oates... or Fleetwood Mac, or Steely Dan. Then some form of 80's throwback. But as far as recommendations go, I really only want to expand in that first genre of Electronic Music. I don't want Jethro Tull to be recommended to me. My feed used to be…

Agreed. I really enjoy the Discover feature, but I wish it could be broken down in sub-categories based on mood.

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

#180

Post by RationPhantoms »

madshiva wrote: post #160

They don't find new music, but music that other already know, then the music is not new but the trend music. It's like recommanding despacito.. all these waste of ressource for acheiveing this is amazing.

What you're saying flies in the face of modern findings of the power of "crowd knowledge" and is actually wrong. I, amongst others, find the feature incredibly useful so maybe it's a waste of resources on just you.
Post reply on HN