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

181–190 of 275 posts

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

#181

Post by gerald42 »

thirdsun wrote: post #47
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.

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.

Yup, I worked on the playlisting service for a little while at Rdio, it was pretty simple but worked well. Based on those presets we would decide how far along the artist similarity graph we would walk from the original playlist "seeds".

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

#182

Post by delegate »

The thing with any 'recommendation' system like Discover Weekly is that while it recommends based on past preferences, the recommendations have the effect of influencing and reinforcing the musical tastes and at some point one will notice that one's taste in music has been entirely manufactured by the algorithm, week by week.

This is not just in music, the filtering 'according to preferences' is ubiquitous in today's applications - so I wonder - were does the recommendation end and influence start ?

For example, Google maps routes you to avoid high traffic, but by doing this, it is also generating traffic and the more people use it, the more influence the app has in the real world traffic.

I for one use it sporadically; my music tastes are so state-dependent - sometimes I want ambient music, sometimes I want heavy metal, sometimes I want lyrics and sometimes I want a hard electronic beat. The algorithm does not know my current state, wether I want to keep or change it - even I don't always understand exactly what and how I feel.

Also, I've had it happen lots of times - sometimes I listen to a track or album which I don't immediately like, but then it grows on me and I discover something beautiful hidden in it. There's value in listening to things that don't follow the usual pattern and that's very hard for an algorithm to do.

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

#183

Post by Jackim »

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…

It's a bit of a nuisance, but you could go into Private Mode[0] when listening to stuff that you don't want to be part of your recommendations.

[0] - https://support.spotify.com/cz/using_spotify/the_basics/how-...

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

#184

Post by sumitgt »

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…

I prefer how Google Play Music does suggestions and wish Spotify could adopt some elements of the same.

Eg. GPM does suggests things like: 1) Looks like you are at work, here's some music without lyrics for concentration. 2) Looks like you are at home, here's some relaxing music.

.. and so on.

Also Spotify sometimes gets things too right, and it stops me from discovering new music. GPM does a better job introducing noise into the suggestions.

However, the catalog of GPM is smaller than Spotify. Cannot find famous songs like "Hotel California" on GPM for instance.

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

#185

Post by whatok »

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…

Google Music has location-based recommendations. Not sure why Spotify couldn't incorporate something similar.

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

#186

Post by hbosch »

mediocrejoker wrote: post #171
mulletbum wrote: post #151

Earlier quoted context omitted.

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.

You end up in a echo chamber and quickly exhaust your recommendations.

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

#187

Post by CabSauce »

delegate wrote: post #182

The thing with any 'recommendation' system like Discover Weekly is that while it recommends based on past preferences, the recommendations have the effect of influencing and reinforcing the musical tastes and at some point one will notice that one's taste in music has been entirely manufactured by the algorithm, week by week. This is not just in music, the filtering 'according to preferences' is ubiquitous in today's…

I suspect that the more successful recommendation algorithms do encourage variation. If I were doing this, I may want to suggest some songs that we're confident that the user will like along with some songs that the user may like based on one musical attribute and not the rest (e.g. ambient, but a different genre).

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

#188

Post by unethical_ban »

I took a genetic algorithms course in college. I was undergrad and the prof needed to "make" the grad course. I recall the idea that when writing these algos, you had to inject mutations into your adjustments, in order not to hit a local maximum.

I think the same of Spotify. Several comments here discuss how it gets too focused, or you're unable to reset preferences, etc. I too wish I could "reset" my daily mixes, or else change it up a bit. Wouldn't even be great if you could have some sort of advanced option to adjust the "genre variability" of your stations?

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

#189

Post by charlesdm »

whatok wrote: post #185
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…

Google Music has location-based recommendations. Not sure why Spotify couldn't incorporate something similar.

Does location matter when dealing with music taste? Just because I'm in Nashville doesn't necessarily mean I love country music?

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

#190

Post by nappy »

eeeeeeeeeeeee wrote: post #179
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.

Spotify does this! Check out the "Made for You" / "Daily Mix" playlists.
Post reply on HN