Announcements
The Spotify Stars Program: Celebrating Values Week!

Help Wizard

Step 1

NEXT STEP

Release Radar includes wrong artist with same name as desired artist

Release Radar includes wrong artist with same name as desired artist

I constantly find my Release Radar recommending songs by an artist (say A) with the same name as an artist (say A') that I might actually wanna listen to. This is extremely dumb as an issue because these two artists are listed as genuinely different artists in Spotify and the newly recommended song by a wrong artist is listed as a song of A' in the system. A reasonable conclusion is that at least Release Radar does not look into the artist IDs but just merely refers to their names. This happens to like 5 different artists to me and my Release Radar is contaminated by songs which I have absolutely no interest in. I believe this is a very basic bug that can be fixed in like 5 minutes.

Reply
319 Replies

Thanks for all the amazing breakdown Mihail.

The architecture limitations I completely understand and really shed a lot of light into how these things can't get captured in time (generated 100s of millions of RRs ahead of time for 0 delay come Friday).

Usually my issue lies with the third point you make "I get tracks from artists that are named exactly the same as ones I follow and enjoy, but it turns out the track is unrelated to that artist and they only have the same name."

If in fact these same named artists point to a different artists page, is it safe to hit the button "I don't like X" without it affecting the algorithm, making it think I don't like the artists I follow?

I constantly, almost weekly, see songs with the artist DEVO (I Follow spotify:artist:0UKfenbZb15sqhfPC6zbt3) but when you click DEVO it points to this DEVO spotify:artist:1VodUBUid6WspbE91YTuo6. I guess my question is, if I press "I don't like DEVO" will it know that I am referring to the one I don't follow, DEVO spotify:artist:1VodUBUid6WspbE91YTuo6 ?

Unfortunate that this issue will persist indefinitely, however.

Thanks for the all the help so far.

Here's a real case of "I get tracks from artists that are named exactly the same as ones..." in some detail. There is an artist called Focus which has nothing to do with the classic Dutch band, but some electronic music group (or individual) instead:

 

Spotify_2021-03-22_09-58-13.png

 

Not bad, in fact (to my taste, at least), but this release is mixed up with the Dutch Focus band. Easy enough -- spotify:artist:0ifzzRKdmtgaHy9cfnnyCR will show this new release:

 

rubs_0-1616418021320.png

On the other hand, if you go to spotify:album:3wdCtc8eL9i4qYB7qeIJYq (the EP above), you'll see the original Focus band in "More from Focus".

 

So this will not be solved ever?


@Mihail wrote:

 

First, some data facts!

 

Since we asked for your reports a month ago, we received a total of 61 reports.

 

...

 

For the 1 month investigative period, that would be 1,360,000,000 Release Radar playlists generated.

That amounts to reports for 0.00000004% of the playlists.


I want to open this up by firstly saying thank you for giving us a better insight as to why this happens, and why it's so difficult to fix.

 

Your figures are not correct though, and I want to clear them up to give everyone some perspective.

 

  • Firstly, you're relying on data pooled from two completely different groups: A very niche subset of users of Spotify (ie. the users of the forums), and comparing their data against Spotify's entire userbase which is unfortunately a bias. Never mind the possibility that not all people following this thread will even reliably submit a report. Never mind the fact that this thread is already years old and surely some posters have left, or given up, moved on, and so on.

 

  • Secondly, that figure of 0.00000004% is slightly incorrect anyway. You've calculated 61/1360000000 which does equal 0.00000004, but that figure must be multiplied by 100 to get a percentage of 0.000004%.

 


And we also know there're reports submitted elsewhere and maybe even a lot of people don't even bother, so let's boost that number by a significant amount - let's make it a factor of 100 000. 

 

...

 

That still amounts to 0.004% of errors.

 

  • Thirdly, with figures corrected, 0.000004% x 100,000 = 0.4%.

 

 

Let's dissect these numbers and put them into perspective.

There're roughly 340 million RRs generated each week.


 

0.4% of 340,000,000 = 1,360,000 users that have been affected by this. Or 620,000 paying subscribers according to data from here: https://newsroom.spotify.com/company-info/

 

 

I understand that these are all just rough figures and there's no way to know exactly how many people this affects but those numbers were just incorrect and I felt they unfairly downplay the frequency in which this happens.

Again, it's good to know what the reasons are exactly that could lead to these situations though.

Hey folks!

 

Happy to see all the appreciation for us trying to clarify things. We're giving it our best!

 

I'll be taking some leave soon, but wanted to jump in and reply before going off.

 

@w9t6l-t6s_ok - If by tag, you mean add other artists as featured, then - yes, you can report them the same way. Please note however, that it's more difficult for the team to discern if artists have maybe collaborated or been sampled or similar. Those would be valid feature reasons. In any case, reporting functions the same for those as well. You raise an interesting point with your 2nd question. As far as we understand, there shouldn't be an impact as the RR focuses on who you follow. To be safe, you can always just pick "I don't like this song".

Something curious regarding your last point about report incorrect metadata: We used to have something like this in the form of a specialized website called Line-In, where users could help correct metadata issues. Due to a variety of reasons we had to shut it down, but it could be something that we look into again in the future.

 

@jeremycorbett - Trust us, we know the feeling when a detailed reply gets flushed down the internet tubes. Sorry to hear that happened! Regarding your post - we can only take action based on empirical data. When we're dealing with numbers that range into the hundreds of millions, gathered by actual usage data, a valid statistical assumption would be that if this was more prevalent, there would be more... well... more of anything really. One thing we'd like to make clear is we're not trying to downplay this, but just to bring it into perspective as to why this is difficult to tackle right now.

 

@arcadechan - Glad to hear you found the info useful! To answer your question - yes, if the two artists have distinct artist pages, that meas they have unique identifiers, that the system uses for all algorithms. That being said, you're probably getting this consistently, as this artist's distributor also seems to upload it consistently with wrong metadata, so it might still happen because during the Release Radar capture, it would have the other ID. Wish there was anything else we could do to help at this point, really. Thank you so much for your friendly tone and understanding!

 

@rubs - seems this has since been corrected and the EP has been separated to its unique page. If you have reported this, it might've been your report that accomplished this 🙂 For the time being this indeed is not something we can prevent from happening, but we are strong believers in the saying "never say never".

 

@spotix - You are absolutely correct! I hope my primary school math teacher doesn't see this. Please don't tell, I received enough mocks from my team mates already :'D I've went ahead and edited my silly mistake, thanks for pointing it out! However, do consider we're still comparing something withing the decimals of a 0 to something within the decimals of a 0. As mentioned to jeremycorbett, we would still be expecting to hear more about this, especially from our subscribers, if those figures are really way off, hence why we picked a really big factor multiplier. Changing that multiplier has an exponential effect on your calculations as well. At the end of the day, we can only act based on factual data and reports. To re-state, we're not trying to downplay this, but rather to provide perspective and transparency. Your last sentence is reassuring that we hopefully succeeded at this. Thank you!

 

Hope I managed to cover everything. Take care all!

Mihail Moderator
Help others find this answer and click "Accept as Solution".
If you appreciate an answer, maybe give it a Like.
Are you new to the Community? Take a moment to introduce yourself!

@Mihail

I just want to point out some other considerations when trying to gauge how many users are indeed affected. The first thing is your baseline, the reports collected over the month. If I understand correctly, in the end you took all reports into account and therefore I assume your extrapolated number is supposed to represent not only the algorithm actually failing but also gauge the amount of manual errors when artists upload the music. If that is the case, the number is already skewed by explicitly saying you only wanted reports of artists showing up in the RR without their own page. I alone could have easily inflated the number of reports without that restriction in place.

The other things to take into account would be how many users actually listen to their RR to begin with? How many artists do those users follow on average? I for one follow a lot of artists (around 1400) and the RR is actually a good incentive to do so (I really like the RR otherwise I would not bother writing anything and its current form is preferable to no RR, to make that clear) but the more artists you follow the bigger the chance you encounter the phenomenon (and how unique the artist names are). The less artists you follow the less chance of a wrongly linked artists there is and the chance that a recommendation is wrongly linked and it actually is spotted as such is even lower because the user does not know what the artists is supposed to sound like in the first place (whether he likes it or not does not really matter in that case). Therefore I would expect the number of artists you follow to actually be the main factor that affects how frequent the issue appears. Of course there is also the question how many users actually use the community.

Those are the factors I can think off the top of my head which affect how likely you encounter the problem.  I'm pretty sure that taking all those factors into account would make the current extrapolated number less generous than the factor of 100 implies. I would actually argue that the estimate is affected by so many other factors that I would trust my own RR as a better indicator of the issue despite its awful sample size.

All that said, I like that you looked into it and made it as transparent as you did. But I don't think that it is not a widespread problem if you fulfil the requirements for it to be a problem, its probably more that people just don't care enough/accept it or simply just don't notice.

 

EDIT: corrected some typos and some rewording for better clarity

You do have a report deceptive content link on artists: https://support.spotify.com/ca-en/content-policy-reporting/

Can you add something there for metadata?

 

Here is one from my RR today: spotify:track:0OXEbTAvGW8oOC0jxag5gt

Tagged Justice which leads to spotify:artist:1P7LEMD31jxiljPh53dax6 I follow spotify:artist:1gR0gsQYfi6joyO1dlp76N Pretty obvious that Justice was not sampled.

Just got recommended a new song from (ford.), but it's someone else with the same name.

@rollulus

 

Hey, I was just about to post the same re Darkside. You beat me to it. 

If I go to the artist page of Darkside from my Release Radar it takes me to the artist I’m not following, which also clearly shows. 

149134A8-3CE0-4E28-8078-C3C26A17A379.jpeg


And when I go to the Darkside I am following it clearly shows just that as well. 

B4E5F374-C90D-40F4-8EC3-40E1834249FB.jpeg

Yeah I was super excited to see that they had a new release. I don’t get how this fits into the “wrongly submitted information” story though. The two darksides are distinct artists in the Spotify universe too. Also, to me again this shows that it’s fraudulent on purpose since even the stylized allcaps is copied.

Thanks for the fantastic community outreach Mihail, 

but wait… hold up.   I carefully read your post and still don't have clarity on a very important distinction between two issues:

 

1. RR shows an artist uploaded a track to Spotify using an established artist URI, either accidentally or maliciously. This could be either the 'main' artist or a featured artist.  This is generally a human issue, is not a fault of the code that generates RR, and can be addressed with the 'Report content' form.  

2. RR shows a track from a new artist with the exact same name as an established (followed) artist, but the new artist has a completely different URI from the one I follow. 

 

#2 is most frustrating as it feels like a solvable problem.  A quick example from this week's RR.  An artist I DO follow, "ford." has a URI of spotify:artist:7ItbAZITSFxSy5LJChXe18, 

but my RR erroneously shows a track from "ford." I do NOT follow, spotify:artist:2wDReEUohu3jvfUSONAZnX

 

Did this incorrect "ford." (the one I do NOT follow) upload their music with their new/unique artist URI, or did they upload it with the artist I follow's URI, and then it was corrected by humans between the time the RR started being generated and when it was finally shown to me days later?  

 

It sounds like you're implying the latter is what happens. If so, could the script do a double check, before the RR is released to users, to check if URIs have changed between when it was generated, and when it was released to users?  I have no clue about the technical debt for such an operation.  

 

If it's correct that "ford." was corrected by a human between the time the RR started generating, and when it was released to me, then the "Report content" form is useless here. The issue has already been corrected, the new artist has been updated to have a unique URI.  However, I am still experiencing the pain of the issue, as the RR was generated against the incorrect URI. Nothing I can do as a user actually helps solve the issue in the future. 

 

The big underlying issue here is the lack of trust/verification system. Instead of fanciful ideas of AI listening and identifying between two artists of the same name, artists should be given a unique public/private key pair during onboarding with Spotify, and the private key is required for any and all releases to their URI.  This would completely eliminate both #1 and #2 issues as described above. 

 

 

 

This still persists. The first third to half consists of clone artists that share the name of artists I actually listen to, but that I have never listened to nor have any interest in listening to. The second third is re-recordings of classical pieces I'm already familiar with. If I'm lucky I MIGHT find one new track I actually enjoy. As far as discovery goes release radar is completely useless in anything other than being a source of annoyance its current form.

Yep, still happening.

I have 1 song saved from this artist.

OverWims_0-1617546505927.png

Now my release radar is recommending me this

OverWims_1-1617546554293.png

Which is from this random artist that just happens to have the same name.

OverWims_2-1617546577451.png

I have zero Chinese/Japanese songs in my library. None. So clearly this is not the correct artist. I mean yes they could sing in both languages but clearly, every song they make is not in English. This issue is annoying and needs fixing. It's not as annoying as when the release radar has remixes for songs you already have or acoustic versions etc but it is annoying nonetheless.

 

Every Monday I seem to be here again... Here's proof there are fake artists. Look for "JOELSON O REI DO SOM AUTOMOTIVO" (just like this, all caps -- means "Joelson, the king of automotive sound") and you'll find several tracks, each showing a different "tuned" car. In my case I'm certain the track shows up because of the legitimate band Moloko, from which he stole a sample. So he (the guy named Joelson) found a way to advertise his modded cars for free... just create a fake artist on Spotify! And you (Spotify) won't do anything about it. Unbelievable.

Spotify_2021-04-05_16-59-03.png

Glad I found this thread, because this is getting beyond frustrating. Every week for the past ~3 months, my Release Radar has contained 5-10 songs by misattributed artists, or people claiming to have collaborated with artists I follow. Selecting "I don't like this artist" has no effect, as literally the same people show up again week-over-week.

 

Here are some examples from today:

https://open.spotify.com/track/34HrXJ8L0wWXyQSLhd305d?si=l79uIz8rQRyRzzNT6i1XEA

https://open.spotify.com/track/12yBFtaEZVyNMEira5wGXL?si=tybvfY7gTfyHKp_uLAlCwg

https://open.spotify.com/track/3urrycg0H2Q5HhaX22Auxl?si=rR7WeKmiTz6G8TRwxI_R_Q

https://open.spotify.com/track/56Y3NI0kN3g4hwOcyBBGl2?si=rI-NcokWQmuk-4t3rZ4Pbw

 

All of these list wildly more popular performers as secondary artists, while I'm pretty sure they had nothing to do in producing these songs, i.e. it feels like people trying to spam their way into playlists. Please find some way to be more rigorous about this, Spotify.

I reported a problem from this weeks RR, where as usual a guest artist on a hiphop song was incorrectly linked to a band I listen to.

 

I report it as a broken track, per the instructions. As I do 1-4 times every week now.

However, this time the response from Customer Support is that they will no take any action since I can't provide the link to the correct artist. An artist I don't know anything about who is set as guest artist on a track from another artist I never heard about.

But I can tell the difference between hip hop and Old School Death Metal by just using my ears.

 

So the message from Spotify is that they don't care about your experience anymore. 

And they want their customers not only to report the errors, they want you to do their work as well. This is frankly unacceptable.

I've been falling behind on this thread. I thought there was some progress made shortly after my last post a month ago or more. The number of inappropriate songs decreased by at least half and some of the remaining ones could be legitimate naming conflicts. However, the posts from others show that there is still significant issues here and the inappropriate selections on my own account have started to increase again.

I've received another Spotify email - "New music from artists you love..." and it's another Rap artist. I don't listen to Rap and the only Rap songs touched by my Spotify account are the fraudulent suggestions made in my Release Radar, which I have spent significant time marking as "I don't like this song/artist".

Why Spotify, why? Why would you send me this email? Dozens, of rock/metal artists I do listen to have released new material in the past month but you send me a Rap artist I have explicitly disliked. Could it be that Spotify is being paid to spam this content to us? Why is the Release Radar the only Spotify list that ever contains these inappropriate Rap artists? I've been paying for a Spotify subscription for close to a decade but the algorithm (and only the Release Radar algorithm) still can't figure out what I like. It sure looks like part of the Release Radar algorithm is to include wildly inaccurate selections that pay to be placed on this list.

Yeah this is still a huge problem. I report several songs each week for incorrectly featuring artists with that support form, and half the time I don't even get a response. And why is it that ALL of these songs with fake artist features are some kind of hip hop or rap garbage? Seems like us being able to block entire genres would help immensely here. A report track option on the actual app would also be very useful.

same here. Spotify if you are reading this, shame on you, for letting us hang like this. I used to report to this form, but now the stopped doing that as well.

Today after another HipHop Garbage artist with the name of Pantera (you guessed it, after the famous metal band), I've had enough. This is getting out of hand!!

 

this used to be the form:


https://docs.google.com/forms/d/e/1FAIpQLScdXRSzYb5zXmHR2dCcpMtEz_t-gHK8ZlVXqT-s0nYrCtU2Ow/closedfor...


Suggested posts