Hacker News new | comments | show | ask | jobs | submit login
Spotify’s Discover Weekly: How machine learning finds new music (hackernoon.com)
551 points by kiyanwang 11 months ago | hide | past | web | favorite | 269 comments

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.

They do have a dislike button though, the thumbs down icon next to the play controls (on the desktop version of Spotify).

It's there only if you listen to a radio, not to a playlist.

There's a way to consume Discover Weekly as a station instead of a playlist. ;)

How exactly? — Ah, figured it out:

- Go to the DW playlist

- Click the "..." options button

- Then to "Go to Playlist Radio"

- "Follow" the DW playlist radio

- You can now up/downvote songs

Wow, what a great user experience.

If you had it all on screen you would talk of button fatigue. Users are the worst.

Did you know Spotify also has a its approve and disapprove buttons in the desktop app on opposite sides depending on what particular radio-like feature you're using?

DW playlist radio != DW playlist

Sometime within last month it appeared for discover weekly too, at least on mac client - https://imgur.com/a/QJluP

Not for Discover Weekly they don't. Only your Daily Mixes and other radio stations.

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.

EXACT SAME THING, happened to me I considered to create new account just because of this....

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

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 Keen, and James McMurtry radio stations because they all play the same set of songs that I got sick of months ago.

I don’t know if it’s a licensing issue—do they save money by keeping song variety down?—or is it an algorithm problem? I wish they would fix it because I’m about to bail for some other service if they can’t.

Pandora is the gold standard for good variety and new artist discovery as far as I’m concerned.

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.

With Pandora I can give it one song, and get a playlist out that won't repeat for a couple of hours (I really wish it was longer!)

With Spotify, I enter in a song, and get lots of stuff from that same artist (something Pandora can't do!) and then after that's done, a bunch of repeats.

Then again my problem may be that I'm trying to use Spotify the same way I use Pandora.

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.

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.

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 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 makes me wonder about the motivations. Did the IP for this song recently change hands?

(I still get the shivers recalling the month when people kept gleefully playing me “Temporary Secretary” by Paul McCartney; a new discovery from their Discover Playlists. I could have marched to Stockholm to strangle a data scientist.)

>“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.

It was also used in “Breaking Bad”, in the scene where Walter sets “Ken Wins” BMW on fire. Thomas Golubić was the music supervisor for BB and Better Call Saul and these shows’ playlists make superb additions for Spotify listening, in my opinion. They’re chock-full of great discoveries like Darondo.

It doesn’t surprise me that “Didn't I” comes up in many playlists, but it does surprise me that the other great songs on that album don’t get worked into Spotify stations. This is a great example of their algorithmic problems.

Good context, thanks both!

> subsequently feel let down upon realizing that they’re (currently) common assignations

This is a really foreign concept to me. Why would you be let down because other people listen to the same song?

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".

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.

Tell me about it. A large part of the community we had at rdio used to meet at in a post-rdio slack channel after the service went down. For months we discussed alternatives and how mediocre and barebones they were compared to rdio. Most of them, including me, settled for Spotify, but the community is pretty much gone.

On the other it led me to double down on my purchased offline music collection, which I already maintained long before rdio.

You (and Spotify's team) would like to read this paper; I wish eveyone making recommender systems would.

Recommender Systems for Self-Actualization https://dl.acm.org/citation.cfm?id=2959189

I'm a former Last.FM user. They'd play a mix of past favorites and related recommendations, then adjust based on feedback (e.g. likes). There was a noise slider that would influence the ratio of new songs (and indirectly bring in recommendations from farther away from your core tastes). When I first found out I could be more adventurous, I was very excited, as I (like most people) pride myself on being curious. As it turns out, constantly discovering new songs from unusual genres is exhausting. This turned out to be a feature I very much wanted, but felt unhappy using and disappointed in me not using. Maybe this is common enough that the Spotify PM for Discover Weekly experimented and decided not to ship it.

It sounds like having it as a slider you can enable at-will is the perfect compromise for this particular feature, no?

If you're looking for more noise to get you out of your music comfort zone then check out JQBX (https://www.jqbx.fm). It's a social music app that let's you DJ and listen to music with others in virtual rooms (similar to turntable). It plays all audio through Spotify so you get a huge library, can save tracks for later, and the random plays can seed your new discover weekly playlists (or you can use private mode so it won't).

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.

Or even just featured in a single / best-of album from the same artist. I'm fairly sure my "library" now contains a bunch of duplicates due to this.

I get constant metal versions of pop songs. Winterplay's Billie Jean was recommended to me for weeks on end. This week I have Turn down for what meets metal.

Seems like a small tweak to let us tell them we dislike this style of music. According to the article a -1 should be able to be put in that matrix (if that's actually how it works).

Also, I dislike that I lose music after a week. Which means I've lost a few good songs from their churn that I forgot to save.

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.

FWIW I find probably 1/3 of my music through DW, 1/3 through manually browsing related stuff on Spotify (if I discover a new artist and see they were featured on a compilation, I might listen to that compilation) and 1/3 from real life (friends' recommendations, unknown bands from festivals etc). For my pattern of use, DW is varied and absolutely amazing. I only listen to my DW probably 1 week in 4 though.

I have the same problem. And we it every fad machine learning thing. It never works for me but a lot of people are happy. Does anyone want me as a test subject?!

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 times, skip others and so on? It looks like by considering these simple points it can be improved by a large margin.

Yeah, I was surprised when I read the start of the article. On Mondays, I will start the day by playing my discover weekly playlist and I'll usually have it running throughout the day, saving 1-4 songs from it to my library. Sometimes none.

I just checked and this week I didn't even save a single song for example.

I guess it doesn't help that I do listen to really different types of music depending on what I'm doing. If I'm at home, I'm not listening to the same thing as if I'm at work (developer). When I'm reading, I'll also be listening to something different from normally. I wonder if that makes my discovery much more "basic", since it's a mix of pretty much everything.

This is why they have Daily Mixes, which are clustered by genre.

Discover Weekly seems to have figured out that one of my most listened-to Daily Mixes is post-rock, so it only gives me that. I'm fine with that. I wish they would push the Daily Mixes and Discover Weekly more prominently in the app interface rather than giving me TOP 10 BRAZIL playlists when I open the app.

Listening to sufficiently different types of music destroys discovery functionality for me. My weekly discovery is almost always filled with music that spans from 'boring and unoriginal' to 'heard this before'. I want some novelty damnit.

I think that one reason Spotify Discovery works so well for me is that I really only use Spotify to listen to one genre of music (however you want to describe it, but it ranges from trap to alternative R&B). Otherwise I'm listening to EDM on Soundcloud. So I end up with a pretty "pure" Discover Weekly playlist.

I don't really use Spotify. But this is why I like Google Play Music, because the playlist suggestions are categorized by what you're doing (rockout at the gym, christmas, working-chillout, working-fun).

But honestly you can't beat just going to youtube for a live stream if you want some chillout-ambient music while you're programming. There's so many talented dj's out there culling new music.

Spotify has playlists for the same things you listed (minus maybe christmas, but I'm sure they do). I've seen the others though listed.

You can follow and listen to other people's Discover Weekly mixes. I listen to Tobias Van Schneider's Discover Weekly more than my own.

I have at least one song I like in every Discover Weekly edition, sometimes several. However, interestingly, the overall quality and interest match varies heavily from week to week. Some weeks are fantastic, others rather lame.

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 enjoy, my music taste seems very obscure.

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.

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 listening to Jazz, I usually hate Jazz. Should it start recommending that to me now?

I can imagine what works brilliantly for some people will be abysmal for others (but I guess a decently advanced ML system should detect that and try different approaches for different people).

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.

Mine is pretty good except for it thinks I want to listen to like ten Scandinavian language songs in a row.

Although now that I think about it, one thing that seems to lead to wacky recommendations is sparse data. If I check out Japanese enka albums the "similar to this album" information has stuff like Justin Bieber, which mostly tells me they have so little data on what kind of listeners enjoy the music that it's pure noise.

This is my experience as well. However, the first song I ever saw on Discover Weekly was this amazing gem, so now I kind of give Discover Weekly a pass for the rest of its behaviour: https://www.youtube.com/watch?v=f2cGxy-ZHIs

The algorithm doesn't seem to work at all for me as well. There's usually just a single song that I actually like in the discover weekly list every other week.

Same experience on my end. I mean, the recommendation system is better than nothing, but I very rarely find something I really like based on it.

Completely agree with this and related posts.

I think curation is massively underrated - importantly, curation allows a filter on quality. Lots of popular music is popular for reasons other than its quality, and Spotify is unable to detect that. Critics can.

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.

I have this problem also, I know of private session and afaik it is not supposed to be used to hide the kids music from algorithms, rather the other way around: hide explicit music you listen to from your kids and coworkers etc...

I wish that option were more easily accessible instead of being hidden away behind multiple menus.

It's right there in the dropdown beside your name in the desktop app.

I don't always keep forever tracks found on discover weekly, but I usually enjoy most songs that are on it when I am in the mood for new things.

Same. After I read this article I wondered if I'm doing something wrong to generate the playlist.

It works wonders for me, to be honest. At least 50% is on spot, almost every week.

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...."

Don't you have Daily Mixes? These are smart playlists created by Spotify organized roughly into genres, and you have can up to six - depending on how diverse your listening habits are. Right now I have:

1 Electronic, 2 Pop, 3 R&B, 4 Indie, 5 Rap, 6 Country/Americana

Note they aren't named like this, I did that. Spotify identifies them only by a number and a few of the artists contained within. You can find them in the app on your Home screen or under Your Library > Your Daily Mix.

These solve the problem somewhat, but, in my experience, they only play songs I've already listened to. So listening to songs outside the genres I want recommendations for will still ruin Discover Weekly's usefulness as a discovery system.

I used to love Discover Weekly, and found a lot of great music through it, but it's gotten less useful over time. A few things that would improve it:

• Multiple Discover playlists sorted roughly into genres, like the Daily Mix but only containing never-before-heard music

• Treat Discover Weekly like a radio station, with thumbs-up and thumbs-down buttons

• Once I've listened to a track a few times, stop suggesting it. At this point I can usually predict what my Discover Weekly playlist will be: all of the songs that have showed up on it before, but that I didn't like enough to click "+" on.

They intersperse some new songs between my already saved songs.

I prefer them for exactly this reason - I’ve learned that I actually prefer less novelty in my playlists than I previously assumed. Spotify seems to understand this, and incrementally alters my playlists, moving through tracks and doing a good job of removing tracks that I skip continuously. I’ve noticed these playlists actually do a good job at discovering music for me within each genre

I find that my daily mixes only have songs I've already saved to a playlist or my songs.

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.

I have three main criticisms of Spotify, just in case there's a Spotify PM reading.

1. I hear repeat songs in my discover all the time, this isn't ideal. I curate my own playlists for moods and know that i'm getting when I play those playlists, surfacing songs I've manually curated in my discover doesn't add value to my experience.

2. I wish I could more easily surface the discover playlists of my friends or those whom I follow. I know that my friends like similar music but the small differences might provide insightful curation data and help me discover new music, so, it'd be interesting to see the data on discovery curation through social connections.

3. There's some chatter about daily mixes here. I find that my daily mixes don't change enough. I'll try them once, two times, then I'll churn from using that feature.

>1. I hear repeat songs in my discover all the time

Something is wrong. Discover Weekly is supposed to have only songs you have never listened to, even once. That's how it works for me.

The case I most often encounter is "oh, you liked x? I'm sure you'll like every remix of x for the next month of your discovery weekly!". Technically not the same song.

I get repeat songs now and then. Not a lot, but a noticeable number.

Perhaps you've listened to the regular version of a song, but the Discover Playlist has surfaced its "radio edit" or some other very, very similar version?

Most definitely not. Same songs. Not referring to radio edits (though I have seen them too, sometimes).

Just want to reiterate gripe #3. The Daily Mixes were, at first, the Best Thing Ever. But that was like a year ago, and they don't seem to have changed at all since then. I've listened to a lot of different music this past year, but keep hearing the same old songs on all the Daily Mixes. I'd expect them to update in more overt ways as my tastes evolved.

>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.

Spotify does this too. Under Browse, many of the suggested playlists are context-dependent and will change depending on the time of day.

You could try this tool from hate5six: http://hate5six.com/sage - historically, they've only worked with hardcore / metal music, but this appears to be a new tool which expanded out to 200k artists across all types. I managed to find 3 folk artists which I am loving, and had never heard of from either Google or spotify. Curious if it is broad enough to work for your taste in music.

(not affiliated with them in any way)

He also did a really in depth blog post about how it works. I haven't gotten around to reading the full post yet, but from the little that I've read he goes into a deep dive about how it works. https://medium.com/@hate5six/sage-an-artificially-intelligen...

I'm very similar- look into the Daily Mixes that are created, this solves that issue for me.

Right now I have six different mixes and each one of them seem to be a grouped into a different genre (or sub genre).

This is the problem I have had with it. Last month I decided to check out some Latino music from Columbia after a recommendation, and my Discover Weekly playlist after that started to get populated with almost all Portugeze songs and Brazillian artists. I did find some great songs in them too, but I would have preferred more Latino music. It was disappointing that their model is not able to properly disambiguate between two popular, yet distinct languages/genres.

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-...

The importance of a good Discover Weekly has caused me to be super 'careful' with what I listen to. I have a separate Spotify account on my family plan where I listen to stuff I don't like. For instance, it's connected to my Sonos so guests at my house can listen to whatever they want.

Something like an "Incognito Mode" would be a good idea for Spotify and similar services.

They actually have one. "Private listening". But you can't assign certain devices to stay in that mode.

and exists: WIN: File/Private Session Android: Your Library/Settings/Social/PrivateSession

This will cause your songs not to be influencing the AI curated playlists (like Discover Weekly and Daily Playlists)

I actually want the opposite - I don't want the stuff I listen to all the time to be recommended to me, I want some of the one off stuff I listen to to appear in the feed at around the same frequency.

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?

Music you listen to at the park may be different from the music you listen to while programming at work, for example. So yes.

It's more contextual to the user. Music suggestions for when you are at the office. Music suggestions for when you're at home etc...

Yeah, I definitely wish they had filters for genres, or more tools to let users control the weighting of things.

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.

I have this exact same problem. Please Spotify, fix this.

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 <genre>", containing definitive representation of the genre, "The Pulse of <genre>", containing songs that fans of the genre listen to now, and "The Edge of <genre>", 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.

> You can use this site to find out what genres your favorite artists are categorized as

That's my problem right there: my favorite song right now is a musician's piano cover of one of his own songs. His music is usually electronic, which I don't like, but I love this one song. So Spotify will recommend me electronic music from other artists, which of course does not fit my song.

Repeat for every author. I liked one song from a German musician, and now half the recommendations are German music. While I can see why network relations make sense, I wish I could say "give me a similar song, not a similar artist".

Pandora works song-by-song (and is not bound by genres).

It's also not available outside the US :-/

Alas. (Though, it is available in Australia, if that's any consolation.)

Proxies work.

I gave this trick a try, but only "The Sound of ..." playlists appear in my searches.

Usually on the "Sound of.." playlist there is a link to the Pulse and Edge in the description. Otherwise the genre might be too obscure and a playlist for it not generated. Also, the "Sound of" playlists are stored under the "Sounds of Spotify" account, while the pulse and edge under "Particle Detector", so you might try searching directly in that profile.

Ok, I see the playlists now. Strong vouch for this tip!

I've been using Discover Weekly for a few years now and haven't had that problem. The only annoyance that it sometimes adds a song I already have saved (and therefore have no need to 'discover').

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.

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, and listen to quite a lot of Norwegian music in Norwegian. Norwegian music is also European music, Scandinavian music, Nordic music, etc., and as a result I get music from Germany, France, Finland, Iceland, Sweden, Denmark, among other countries and languages. However, the reason that I enjoy listening to Norwegian music is because I speak and understand the nuances in the language perfectly, while this is not the case with any of the other languages.

In Norway, we have «russefeiring» (russ celebration) (https://en.wikipedia.org/wiki/Russefeiring) from approximately mid-April to mid-May. In the recent years, many groups of «russ» have been making/ordering songs to represent them throughout their celebration, and in that period, I listened to some of those songs because it was fun during that period (around April-May), but it's not interesting to listen to outside of that time period at all, pretty much like Christmas songs. Now it's the second week of October, and my Discover Weekly list still contains a lot of songs (12 of 30 songs) created specifically for the «russefeiring». Imagine getting half your Discover Weekly filled with Christmas songs in May, because that's pretty much my experience with this.

I don't know what they have to do to make Discover Weekly not give me shitty suggestions, but right now it keeps giving me suggestions that are outdated and uninteresting and I have no good way to give that feedback.

I'm really saddened by this, because for the first half year of Discover Weekly, the playlists were so good that I stored them in separate playlists to be able to go back and listen more to them.

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.

I have the same issue with Iron and Wine and I don't get why. Every recommendation engine I've tried decides I absolutely love them, no matter what seed I start with or what songs I like/dislike.

I don't mind Iron and Wine but they are definitely a 'meh' for me, so it gets frustrating when they dominate every playlist. And, disliking/thumbs down never seems to get rid of them.

Are they generic enough that the algorithm finds something similar between them and everything I like?

You might be on to something. Iron and Wine might be the Kevin Bacon of music recommendation algorithms.

My problem as well. I recently was in charge of putting together the playlist for a friend's get-together. They have significantly different taste in music from me, so now my recommendations are all messed up!

We had to build some explicit filters for this at Spotify. For instance we blacklisted Christmas music from being part of the listening data – until them people would get a ton of Christmas music recommendations every January. But to your point, there's a long tail of other contextual things.

Weirdly, some genres seem to be entirely absent. I listen to a lot of jazz and classical, and with the exception of the occasional jazzy pop song, I never get recommendations for those genres.

Granted, Spotify doesn't have a huge amount of contemporary jazz (much of what I listen to are imported ECM tracks that Spotify doesn't have, though not all), but it does have a lot of classical.

I don't think it's ever recommended any Latino or folk music, either, come to think of it.

classical music has notoriously bad metadata on spotify – it just doesn't fit into the data model of artist/album/track. i think that causes the system to filter out a lot of the listening data or fail to find patterns.

not sure what's up with jazz but i'm guessing it could be a similar problem

Maybe they can factor out those event/location-driven trends, or dampen them down.

I think the Christmas example is a good one, I wouldn't want half my Discover to have Christmas songs- but I wouldn't mind having 1 or 2, perhaps from an artist I had never realized had a christmas single.

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.

Thanks, but this is what I've been doing for the last few months, although I have also listened to music that is somewhat similar in genre (Norwegian music, electronic, hip-hop, pop, etc). I don't think I should stop listening to the music I like in order to manipulate the algorithms into giving me the music I like and want to listen to. It's counter-intuitive.

I built the foundation of this system while at Spotify. While it's true that we looked at a lot of different signal, at the point when I left (early 2015), it was all based on collaborative filtering.

The reason collaborative filtering works so much better than anything else is that given enough data, it will already encompass everything else. If there are reasons why certain users prefer certain sounds, or certain lyrics, those patterns will emerge in the listening data.

The main reason to use any non-CF method is mainly for new content that Spotify doesn't have much listening data for.

I'm no longer at Spotify, but let me know if you have any questions

I guess this is a less technical question - but what would your advice be to an artist who wants fans of their more famous influences (similar artist) to discover their music? Aside from the obvious (tour, market yourself to popular playlists etc).

Sounds like there's some NLP and web-scraping involved...so would it make sense to come out with blog posts that compare you to the famous artists that influence you?

only thing you can do to influence the collaborative filtering system is to make users listen to both you and some other artist.

as i mentioned in another reply, CF is really what powers DW

In this context, what does CF and non-CF mean?


Collaborative filtering methods vs non-collaborative-filtering recommendation methods.

Thank you.

Even musicians need to understand a little bit SEO ;) Software is indeed eating the world.

Is your listening history from last.fm taken into account when signed into it via Spotify?

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).

Haha - read your comment, looked at your username, opened Spotify, and "Purification Ritual" is the first song in my Discover Weekly.

Hah— no way! That's amazing... made my morning :)

And now thanks to discussion of Discover Weekly, I've discovered Steve Benjamins... Thanks!

And "Hi Steve!" - listening to "Purification Ritual" now...

Hi! :)

Never thought I would find a new artist on hacker news that I would like. Purification ritual is great!!

That's amazing— thanks!

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.

How do you get onto discover weekly?

Discover Weekly works really well for me. It surfaces a lot of music I've forgotten about and for the most part provides an interesting collection to listen to. Having said that, I have a very eclectic taste in music, so it's probably harder to hit on things I won't like. My process is to try to listen to it several times through and then pick out the stand out tracks once everything has had a chance to grow on me.

One thing I worry about with it though is how much my behaviour might influence the choices. For example, if there's a track in there that I already know quite well, because I like it, I'm often scared of skipping it in case the algorithm takes that as a massive negative signal.

In terms of behavior influence, I listened to pretty much exclusively Hamilton for like a month and didn't get very many new Broadway or hip-hop songs (which is good for me because I tend to not really listen to either genre). So there must be significant weight given to songs/genres you've historically listened to a lot.

Or, based on the way the algorithms work, people that listen to the Hamilton soundtrack a lot, don’t listen to hiphop/broadway a lot.

Same. Also, anytime I get a bad recommendation I blame my daughter for wanting to listen to the Frozen soundtrack.

Time to introduce her to Moana for a change of pace.

Interestingly (and now very off topic), she hasn't even seen Frozen. She does not yet know it is a film AFAIK, and I'm in no hurry to tell her.

I believe that the concept of Frozen has been explained to her by her peer group, through the medium of hair plaits.

Three year old kids are a weird and amazing bunch.

Thinking about it a little more, we've listened to a lot of Moana / Frozen / Disney on my Spotify account and none of that has come up in the Discover Weekly, so I guess they filter for it somehow already. Though I guess when they compare to other people with the same tracks you'll probably get a bunch of listeners that fall into a similar category anyway. Also, it doesn't sound like the number of times you play a track adds a huge amount of weight anyway.

Wrt fear of sending signals, I feel the same.

Also, I have a young daughter who often asks me to play music in the car or Google Home and those affect my recommendations. I wish Spotify had some switch I could turn on to temporarily ignore anything I did until I switched it back off.

What you're after is what they call 'private listening' or 'private session': https://support.spotify.com/cz/using_spotify/the_basics/how-...

Oh I didn’t know that! Awesome! Thank you.

With all that AI excitement Spotify, Google Play Music and Apple Music misserably fail at generating good musical recomendations. At least for me. Spotify Discover Weekly was no different for me when I still used Spotify (six months ago). Not only it is not personalized at all it is also quite lame. I am doing enough to seed - following enough artists, genres, liking album, subscribing to playlists but the quality of recomendations is mediocre at least.

For example at the moment Apple Music is suggesting to me four Wednesday playlist - all heave metal the genre I never liked and listened to. Also two artist spotlights - Jeff Chang and Danny Chan. Yes I have Japanesese account but this is Cantopop and Taiwanese crooner - so a little off the map especially that I do not enjoy pop music at all especially Asian. Some other examples are what I call comin denominators. Yes I follow lots of jazz but Frank Sinatra is not jazz music, nor Tony Benett. etc...

Same here with Netflix: their recommendations have been a complete failure. It seems like Netflix is pushing mostly their Netflix original series. What does works for me in Netflix are the similar to this below the movie, e.g. after you've viewed the movie. But that is a very simple algorithm any sophomore could write, not related to the recent developments in neural nets/machine learning.

When I go to an filmfestival or an concert, part of what I pay goes to the proffessionals who make the selection for me. And I'm happy to pay their services, just like I pay for journalists.

I have the complete opposite experience compared to you. All of my Discover Weekly and Daily Mixes are at least 75% music I like, even if I didn't know it beforehand, or only knew the artist by name.

At bare minimum I really with I could tell these services which artists I explicitly don't like so that they're never recommended to me. That alone would be a huge improvement.

Disliking multiple songs from the same artist only to have it suggest the next one from their repertoire is infuriating. Worse, Apple Music will play songs you've disliked in the radio anyway.

My experience was that Spotify radio is utterly bad, Discover Weekly quite okay, but the weekly Apple Music playlist really nails it. I found a lot so far that I didn't know before and added to my library.

> 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 on an existing playlist and then "Create Similar Playlist" - that gives way more control over what kind of genre/style should the playlist consist of.

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.

There's no reason that I can see not to combine both explicit and implicit rating. If I really love a song let me mark it as such, but also feed things that I've chosen to listen to/skip/add to my library into that algorithm (possibly with a lower weight).

You already can 'save' songs in Spotify. I would hope they take that into account.

For anyone interested in the Raw Audio Models section of that articles, there are some fun endpoints[1][2] from the Echonest API that provide those models.

They've moved over to the Spotify API since I last had a play, but it's great that they still provide them.

You can get the audio breakdown of a track, as well as a summary of the track features including fun stuff like "danceability" and musical positiveness ("valence"). Radiohead's "Fitter, Happier" was low on both of these points if I remember correctly.

[1]: https://developer.spotify.com/web-api/get-audio-features/

[2]: https://developer.spotify.com/web-api/get-audio-analysis/

I also wanted to nitpick this part of the article -- the neural network isn't determining the audio features; these are all hand-engineered features developed by The Echo Nest.

The neural network is trained to mimic collaborative filtering vectors from raw audio. It's a separate model from time signature, key, mode, tempo, loudness, etc.

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 annoys me to no end, same for daily mixes. I end up listening to the same songs over and over and over and over again. Maybe that’s what makes them the most money?

Last.fm was amazing at finding me new music I liked. Rdio was amazing at.. radio :D I used to go for months on the same station. I miss both a lot, and occasionally I still use last.fm or everynoise.com to generate better playlists for Spotify.

Spotify's daily mixes seperate the genres/moods. I occasionaly switch to listening electronic music, but I mostly listen to rock. Spotify has one daily mix of rock, one daily mix of heavy rock and one daily mix of electronic. It doesn't mix the genres.

But discover weekly only gives recomendations for the dominant genre.

>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.

I don't listen to pop very often - that's how I notice it. Eventually I'll realize that a song it added to my playlist is playing in the background in a shop, on TV or the radio.

The rdio Radio was actually very good. You can find a similar approach in Mentor.FM (and you can import your last.fm history to feed it).

A curious side effect of Discover Weekly is that it sometimes influences what I listen to, i.e. I'm afraid to listen to a song in a particular genre because I don't want Disover Weekly to get the wrong idea. But then again, maybe Discover Weekly knows me better than I do.

I had the same issue: I spent a couple of days listening to old tunes I rarely pick anymore, and it “messed up” my Discover Weekly. I thought “Well I only have myself to blame, don’t I?”.

Turns out Spotify has a “Private session” mode to prevent influencing your taste.

Just listen to music. Don't worry, listen to what you like, create your own playlists. Use Discover Weekly for inspiration, not your sole source of music.

Discover Weekly is weird. For the longest time it kept recommending Finish music (roughly 30% of all recommendations), I'm from Sweden and barely ever listen from anything from Finland (or Sweden for that matter).

They have a ridiculous bias towards covers. I bet I've been recommended 50 (I wish that was an exaggeration) version of the Gladiator theme ('now we are free'). And they are all terribly bad (as in blood coming out of my ears bad). I am genuinely ashamed that someone thought that it would be a good idea to submit it to spotify - and if so how spotify could recommend it to anyone, there is no way spotify could have gotten any indication that anyone has ever liked any of those versions of that song. Similarly I get lots of game of throne covers, and now Despacito covers (never listened to that song in any variant on spotify, on purpose at least).

I have a similar problem with remixes from trance tracks I listen to regularly. Some of them have literally 30 different remixes, and they keep getting into the 'discover weekly' playlist. I guess these would be coming from the acoustic profile or the NLP recommendation algorithm described in the article. That said, I still like the feature a lot, it's a little naive to think it could only ever recommend songs you will actually enjoy.

In my experience (listening to ~95% metal/hard rock), all of the automated playlists work amazingly well.

I’m from Sweden too and I always get at least 2 german/danish songs in my discover weekly and I hate it.


Spotify's discovery engine playlists (Discover Weekly and Release Radar) just don't work for me. I now ignore both playlists and I suspect that the lists are influenced by payola.

I've tried training the algo by following artists and saving albums in the style that I would like, but these playlists keep peddling stuff that is way off the mark. Interestingly, the daily mix playlists have responded to this training, but not Discover Weekly or Release Radar.

For users who are tired of the same songs being fed to these playlists each week, you can create an IFTTT action to save the content of these playlists in separate archive playlists. Once a song is in the archive, it won't (or shouldn't) appear in either of the weekly so-called discovery playlists.

edit: grammar

> Once a song is in the archive, it won't (or shouldn't) appear in either of the weekly so-called discovery playlists.

They still do, sometimes, at least for me. But more importantly, adding songs to playlists sends the positive signal to Spotify’s recommendation engine, as far as I know. So, saving all your discover weekly tracks into archive playlists will encourage Spotify to give you similar music in the future.

They don't work for me either. Most of the songs I like in these playlists are the songs I already listened to hundreds of times.

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 Iron Horse


Collaborative filtering seem to be a much bigger contributor to the recommendation right now than objective factors about a song itself.

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 (specific songs). Now take those 1,000 people and compare the other artists and songs I am listening to. Let's say, of the 1,000 people who like Spears and Pantera, 680 listen to a specific song by Creed. That's a high hit rate, which means that song should be offered to me as I will most likely enjoy it. It doesn't matter that Creed doesn't sound anything like Pantera or Spears.

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.

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.

If you're into metal, hardcore, or anything in between, you've seriously gotta check out https://medium.com/@hate5six/sage-an-artificially-intelligen... - it's from the guy who runs hate5six and it's fantastic

I use Spotify's Discover Weekly on a regular basis. On average I like more than half of the songs on the playlist, and it's my main way of discovering new music.

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 subscribers. As a former paid subscriber I can conclude only that they either lack a QA department or I ended up in the A/B bucket from hell.

> combined with not having a way to play album tracks in order

As a paid subscriber for several years, I've had no difficulties playing album tracks in order. I'm not sure what issue you were having, though.

If memory serves there was a button which was supposed to toggle shuffle and ordered playback. That didn’t work on either iOS or the desktop apps and the support team took a week to respond with a copy and paste of the instructions I’d said I was following in my original email.

I’m totally willing to believe that it was some config bug but the effective lack of support was quite a disappointment, especially with Rdio for comparison.

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.

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?

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) :)

I am still missing Rdio in that regard, they were just playing the right songs all the time. With Spotify this is absolutely not the case and it's even far worse than YouTube.

While I haven't found a recommendation engine that works well enough for me, rdio's was very close to it.

It's big feature, among others, was their heavy rotation section - I only followed people that shared a similar taste to mine and we had a really cozy circle of listeners whose current favorites were surfaced by said heavy rotation section. We discussed albums in comment sections, shared playlists and I regularly stumbled upon familiar usernames and friends when discovering new gems. This social component to discovery is completely missing from Spotify, yet it's more powerful than any recommendation engine I have used, including Spotify's attempts, which I'd rate as mediocre.

I miss rdio.

Opposite experience here: Spotify is absolutely perfect. They have conditioned me over the past 7 years to love whatever they recommend.

YouTube has good automatic playlists but the play-next feature often wanders way off after a few songs

I wish I could feed Spotify ML algorithm with my last.fm play history (100 000 songs) or rateyourmusic albums (3000 rated). It would be much more accurate (probably about 3000 plays from Spotify itself)

Rdio worked the best because I could befriend people with similar tastes. I don't understand why this social aspect isn't being used explicitly on Spotify. Rdio knew I had some eclectic musical tastes and yet could find me other people with similar tastes - I found my best music by browsing through other people's collections of music.

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.

Yes, I remember thesixtyone. You're right it was a great discovery tool for indie music. I just browse reddit.com/r/listentothis mostly on the comments to find something unique and interesting music.

They're not anywhere near as last.fm's recommendations.

In my experience they got pretty close to Pandora's recommendations already, which were excellent.

last.fm for me just recommended the next big artist in the same category. Often their music styles were still obviously different and the recommendation rather put me off. People that are hooked by the complexity, details and perfectionism in "Nightfall in Middle-Earth" won't necessarily like Manowar.

I recently got one Spotify recommendation that lead me into listening through a band's full catalogue and getting tickets for their show two weeks later (Insomnium, btw). It also dug out a song that I liked in primary school but completely forgot about. They discovered that I like cheezy metal covers of 80s pop songs and add one by some obscure band to my list every now and then. I'd say my experience was often pretty accurate.

Thanks, that is one of my main gripes with most recommendation engines, and since you like metal, the examples really resonated with me. For me, it was Wardruna - an ambienty, folky acoustic group. Listeners often also like the "typical" viking metal bands, but damn, if I want something like Wardruna, I do not want death metal singing about the same topics, I want, say, Forndom. The same often happened in electronic music as well: if I listened to 5 relatively unknown psytrance songs I liked because they have a nice balance of melodies layered with very distinct synth sounds, then no, I don't want to liten to David Guetta next, because that is also trance-like.

In other words: yes, last.fm's recommendation was much too much based on "customers also liked", which helps in a lot of cases, but so often it horribly fails ("Customers who bought The Martian on Blu-Ray also bought this asthma medication because chance happens. Wanna try it out?").

Yup. Gotta feel sorry for last.fm, they should have been front and center in this industry but let it slip away.

Last.fm's customized radio didn't work well for me - it'd play the same limited selection of tracks every time, often one song from an album and never play any of the other ones on it.

I disagree, Last.fm's recommendations have always felt lukewarm to me. Discover Weekly, on the other hand, has introduced me to TONS of new music that I still regularly listen to.

Last.fm's artist radios are the things i miss the most related to music.

Yes, exactly. I discovered tons of new artists with Last.fm radios. But Spotify is playing same songs again and again and again.

My problem with discover weekly is I basically listen to boring background music like vaporwave all day at work. So while I like a large variety of music from all sorts of genres, since I spent 8 hours a day listening at work, my discover weekly is just slammed full of that stuff.

You could turn on Private Listening[0] at work, that way the boring background music won't affect your suggestions.

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

My Foolproof Guide To Managing Your Discover Weekly Playlist:

- Only listen to the DW playlist once through. Find the songs you even remotely like, put them in a new playlist. Listen to that playlist instead (I call these playlists "DW-{datestamp}").

- Find your favorite songs in the playlist, explore that artist/album. Even if you don't think you'd like the other stuff listening to more of an artist seems to help suggestion variety.

- Don't let your listening for the week be dominated by a good Discover Weekly playlist... every time I do this my next 2-3 weeks are total crap. If you must repeat the same songs over and over, move them to a new playlist.

- Try to mix up genres as much as possible... listening to different genres that aren't your favorite often leads to Spotify recommending off the wall artists in the genres that are your favorite.

- Keep a playlist of your most frequently listened songs, regardless of genre, artist, etc. Whenever you want to listen to one of favorites listen from that playlist, instead of going to their artist page. For some reason this seems to have a large effect on my recommendations.

Most of these suggestions are due to personal experience and two theories:

1) Grouping/organization/total play count of playlists influences recommendations much more than people think

2) It's very easy for Spotify to get into positive feedback loops, forcing variety and constantly curating/making new playlists expands your horizons and keeps the recommendation engine from repeating songs/artists too much.

There seems to be a common misconception, even among programmers who should know better, that Spotify will just instantly and always offer you fresh, undiscovered, music you like on demand... which isn't how these systems work. It definitely makes it easier to find new music, and is a valuable tool in finding new artists or under marketed artists, but you're still going to need to put in a modicum of work curating your own experience to get the most out of the discover playlists.

Feels like an SEO hacker trying to understand and influence a black box search result ranker like Google's. Which is kind of nuts and you'll always be a step behind. Would be pretty cool and interesting if a recommender like Spotify would publish a guide on how to get the most out of their system. Like what you did, but from the builders themselves.

I discovered how good Spotify's Discover Weekly was just a few days ago. Basically, I liked all but 2 songs on that list.

My taste in music is also rather specific, which made it even more impressive. I shall see how the following weeks fare.

It's nice the first few times, but after a while I get the impression that I'm trapped in a "Groundhog Day" loop and hear the same music over and over again.

Spotify should add a slider that lets me widen or narrow the 'search area', sometimes I want to hear more similar music, sometimes I want to find more stuff at the edges where all the interesting stuff lurks.

Spotify seems to mostly stick with the ~5 or so most popular song by a particular artist, so I am seeing some repetition. I don't mind, since I like most of those songs, but you should really take them as artist recommendations, and explore their other songs.

I've found, that if you put in more effort into discovering music yourself, Spotify's recommendations improve.

That option would be useful indeed.

I used to browse the community forums, propose features and vote on others. Spotify, however, has been pretty unresponsive to even the most reasonable and popular proposals, sadly, I might say.

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?

I'm going to have to go ahead and plug SAGE[0], because between Spotify recommending me the same bands/songs ad nauseum (or just being way off), friends already telling me they used it to find a new band to listen to on the first day it launched, and just being a fan of his work in general, it's a complete labor of love and I think it shows.

[0] https://medium.com/@hate5six/sage-an-artificially-intelligen...

Never thought I'd see hate5six on HN. Nice.

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.

Did you read the article?

Only 1 (audio analysis) of the 3 models (collaborative, nlp sentiment, audio) doesn't mix in recommendations from non-you sources, thereby surfacing new music to your attention.

It explains why I tend to like Discover too. Precisely because it doesn't duplicate my exact tastes.

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).

I've been using Apple Music the past 8 or so months because somehow while I was finishing up school I ended up on their student discount plan and not Spotify's.

I really miss Spotify's Discover Weekly. Apple Music (AM) has a similar feature, called "New Music Mix", but it's never as accurate.

While I was still using Spotify heavily last year, almost every week I'd duplicate the Discover Weekly playlist so I could keep that exact mix because a majority of the songs would really fit my current music tastes. Nowadays with AM I only duplicate a New Music Mix playlist once every month, if that. It's ridiculous how uninspired the playlist from Apple is and oftentimes it includes music that isn't similar to anything I listen to.

Discover Weekly is the one feature that would bring me back to Spotify and let me ditch AM once my discount period expires.

Isn’t this article conjecture written by an enthusiastic fan?

Do we know anything about the algorithms actually used?

It's very impressive. But, I'm curious about some of the technical details. For each of these representations you basically end up with a dense vector on a song level. Which I assume you would then kNN with a user specific vector. But I've never come across a nice kNN data structure that supports high dimensional vectors in a larger than memory setting whilst supporting updates. Spotifys own Annoy is cool https://github.com/spotify/annoy, but changing or adding a song requires rebuilding the whole structure ... surely that's prohibitive at scale?

Spotify re-runs the latent vector models regularly and re-indexes them into Annoy indexes. There is no need to do that in real time, you can be a few weeks delayed and it's usually fine. New music doesn't have much data and need different methods anyway.

I read a few threads down but couldn't find it so here it goes:

I'm quite unimpressed by this feature. It knows full well that I almost only listen to music without lyrics (trance genre to be specific) when I work.

Sometimes I try the automatic playlists including discover weekly and what do they play? 50% vocal.

I skip as soon as I recognise it but they never learn. I also mostly play from a precompiled list containing only instrumental trance but even that is not enough.

At least they are in good company. Google has seen all my searches, my photos and my mail since before I met my wife and yet for years they figured out the most relevant ads they could show me was for dating sites :-/

They still suck compared to what Last.fm recommends me. Also the fact that Spotify uses implicit feedback instead of explicit like Last.fm pisses me off. I'm constantly left asking what the algo would make of my actions.

Here's something I would never be recommended. I discovered the group by chance when listening to NPR one day -- The Frightnrs: https://www.youtube.com/watch?v=gwibcfkfuGQ&list=RDEMedGoIEt...

It's really unfortunate that their lead singer died of ALS as the group completed producing its album.

This recommendation was based on the preference of a third party (the DJ). I discovered by sharing his listening preferences.

Personnaly, I still use the good old "This Weeks Releases" playlist managed by user jlarome.


I adore Discover Weekly. No lesser adjective will do.

Now turn the data into a dating/social app so users can discover the cools that share their exceedingly good taste in music.

Does anyone know how they evaluate how successful their new recommendation algorithms are? How much do they improve on simpler algorithms?

It's not obvious/intuitive the NLP and audio model algorithms mentioned would be that successful. I would have thought collaborative filtering + showing you new tracks from artists you like + showing you new tracks in genres you like would get you most of the way there.

Content based filtering methods like the NLP and audio analysis approach can be used to diversify the results that you'd get from collaborative filtering (or add 'noise' to recommendations, as someone up the thread wished for).

I'm guessing they evaluate in many ways. E.g. play counts for online evaluations, accuracy compared to some ground truth for offline evaluations

I really don't understand why collaborative filtering isn't the sole approach. It's basically exactly what you want if you have that volume of data.

Relying on collaborative filtering (using play counts, as Spotify do) can lead to non-diverse and non-serendipitous recommendation. If I like Abbey Road then a typical collaborative filtering scheme might tell me I like Sgt Pepper's and Let it Be. Excellent recommendation, but useless if I'm already familiar with the band and the genre.

There might be an unknown Japanese band that'd suit my tastes perfectly and I'd never know with Collab filtering.

That just seems like a narrow net problem than an issue with the approach. Assuming I have some distance function that tells me who's "similar" to me and it's malleable, I should be able to reach those outliers.

Yeah, makes sense, and for sure works for many cases. But there's no meaningful distance between two sets of preferences that have a null intersection.

I've come across this just as a user with regard to film. I reckon that I've watched all of the 10/10 mainstream comedies that have been made in English. I want more 10/10 comedies, so I'll have to look at foreign languages films. But there isn't enough overlap of user preference, due to demographic separation I suppose, to get decent recommendation from other countries. If you start adding metainformation preferences (e.g. must not be in English, must be >9/10 stars etc) then you're back to Content-based recommendation territory (i.e. audio analysis as Spotify are doing)

Discover weekly is usually just generic garbage. The daily mixes aren't much better, usually playing the same songs over and over in a different order.

I have a ton of songs that I've found over time that I like that spotify has managed to kill for me.

I really enjoyed soundcloud's music and tended to find way more interesting stuff on there. Too bad it's apparently run by morons.

One thing I want in Spotify recommendation engine is proper multilingual support. I have a varied/eclectic taste in English/Hindi/Punjabi music that spans decades and there are many sub-sub-genres that Discover Weekly doesn't capture. It keeps recommending me Telugu/Tamil song, probably based on beats, but I cannot understand them haha

I find that with spotify I get a more consistent variety of music. However I actually pay and use google music primarily. Wish there was a feature on google music (besides feeling lucky or radio) that had curated content like spotify does. I'm usually able to find at least a song I like that comes up in the discover weekly

Weekly suggestions is pretty good for me but daily suggestions can be horrible for less widespread listening. Yesterday I listened some Iranian traditional music for first time with Spotify and It didn't take much time to suggest based on traditional and theosophical music, some dirty hip hop music full of swear words.

My problem with Discover Weekly is it weights recent listens way too heavily. I was listening to R&B for awhile and really loved the new songs it was pulling up - then I switched to jpop and then the list was entirely jpop. Over time, if you don't listen to other music it becomes kind of eclectic / coffeeshop pop.

I think that some people don't like the Spotify Discover Weekly because they don't want to admit that they like crappy music. Its kind of like when you hear your voice on a recording, you're like... "do I really sound like that?"

Just face it, the algorithm has you pegged.

For the first few months, I really liked DW. The last few weeks, it's been giving me almost exclusively country rock. That is very different than it was before, though I would like an occasional country rock song (The Outlaws, for example).

I'm kinda bummed by this, since I really looked forward to listening to it on Monday going to work.

I don't think I'm above liking anything.


From my personal experience Spotify has really bad recommendations, even YouTube has better ones.

Other Spotify features include: your daily mix (like discover weekly, but daily, and providing an individual playlist per each genre), release radar, charts, suggesting songs given a playlist, and creating a radio from a song.

Google Music's "I'm feeling lucky" is awesome. This is very subjective, but I feel it makes much better predictions as to what I'll like than Spotify. Maybe someone else has a different experience?

Google Music was, in my opinion, the industry's greatest music discovery service back when it used to have the "Explore" tab where you could drill down through genres and sub-genres to find popular or classic albums within that genre. It also used to give me much better recommendations on the homepage for new albums that I might like.

It's all gone down hill since they switched everything to mood-based radios. I feel like the main problem is that I like to listen to albums and Google seems to assume I only want to listen to random streams of disconnected singles. I seem to just get played the same stuff over and over, too.

I don't know how to find new (new to me, not the world) music anymore. Are there any good services that anyone recommends?

> I don't know how to find new (new to me, not the world) music anymore. Are there any good services that anyone recommends?

I have the same problem. Usually I find new stuff on the various music subreddits and last.fm's Similar Artists page. Even though I have tons of playlists and saved songs on Spotify, their recommendations usually suck (probably because I listen to just about every music genre under the sun.)

Their radio feature rehashes the same songs I've heard dozens of times, and Discover Weekly is usually so awful I never listen to it.


Applications are open for YC Winter 2019

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact