In the “Release Radar” generated for me last week, there was a 40 second song consisting of rhythmic mechanical noises with a strange title and album art of an alien head. I didn’t like it at all. I think it was recommended to me because the publisher listed its artist as “Fanny Mendelssohn” — a classical composer whose work I definitely enjoy and have probably pressed “Like” a few times on — even though the song had nothing to do with her.
You’d think, she shouldn’t be releasing any new works, you know, on account of being long-deceased and all. Therefore Spotify could block releases claiming her as an artist. Unfortunately, there are new renditions of her work being performed, recorded and published regularly, by many different groups. All of them have the right (perhaps even an ethical duty) to put Fanny Mendelssohn as one of the artists. These are works I’d like to hear, so Spotify’s recommendation algorithms are on the right track.
How on earth can Spotify distinguish real Fanny from fake Fanny?
Wasn’t the whole point of Brexit to have less regulation on things like region of origin? British Fanny is just as good as continental Fanny, even if it isn’t as popular. It’s not “fake Fanny”, it just gets less exposure.
To be fair, Spotify often conflates lesser-known artists that have the same name, particularly if they both inhabit some kind of "alt" genre, even if the two genres are totally different alt genres. Are they two different artists, or did they just change their sound?
Sometimes it's obvious (e.g. techno DJ and death metal band with same name), but sometimes it's not even easy for a human to decide (two unrelated indie bands from the early '10s with the same name).
So it's not too much of a stretch to imagine that it can be exploited for money laundering and/or algorithm revenue farming.
Probably serious enough, since the album they describe exists. Spotify is smart enough to know that it isn't "the" Fanny Mendelssohn though - at least in the artist database. It could well be that Release Radar is dumber and is fooled by the matching name.
Release Radar by its nature will try to recommend tracks with very few plays (if any). It is supposed to solve the "bootstrap" problem in recommendation systems.
In retrospect, I think it's more likely that Fake Fanny Mendlelssohn did NOT inform Spotify that they were a different person, and Spotify only split it out after a lot of Felix/Fanny fans clicked "I did NOT like this" when the track was put in their Release Radar.
The Release Radar playlist is supposed to contain new releases from artists you already listen to. I think you might be right about the release radar being a method they use to separate out bands with the same name. I've definitely noticed bands with the same name in my Release Radar. It would be useful if Spotify exposed a method to report different bands under the same name. The do not recommend is a different signal since maybe I also like the new band. I probably mess up Spotify's recommendation algorithm for me since I use the heart button to flag albums as listened to regardless of my enjoyment since there isn't a another way to mark albums as "listened" like there is for podcasts.
Fanny Mendelssohn was personally involved in roughly 0% of the recordings that come up on Spotify when you search for her name, but I'd bet any fan of hers would be OK being recommended most of those recordings, or most any new ones that came out using her as a search term.
Classical music categorization (where the composer of the music is often of more interest than the performer) is very difficult to reconcile with pop music (where the opposite is more likely to be true).
This is incorrect, but it's principally incorrect because we all abuse the artist field to contain composer information. This is done, I surmise, because figuring out when to display the composer information (performances of composed works) and when not to (bands) is too hard while delivering consistent UX.
And on the file level, it is! But then do you as a music player / streaming service display that all the time, when most people are not listening to classical music and do not care about the composer? Do you make the user manage which columns are visible? (Almost every HN user misses when more software did this. Almost every normal user found it confusing/annoying.) What about publishers who do not give a damn that Johann Sebastian Bach did not perform this digital recording because they know it'll SEO way better to have him in the artist field? Now are you displaying all that inconsistency?
I don't mean "add composer to the metadata", I mean add a field "primary artist key" which could have the value "composer", "performer", etc.
How many fields are displayed is up to the UI, but having metadata to indicate what the most sensible field is would allow a simple UI like Spotify to just show that by default.
Specifying on an individual level avoids trying to lump together genres and categorize which field matters most for each.
Isn't it principally an industry thing to hide the composer, they don't want us to know that the pop is from a "factory" and picked for the artist, they want the public to believe the myth that the manufactured band sit on mountaintops with their instruments coming up with new grooves (or whatever).
Can't you just let the user decide "artist + track" or "artist + composer(s) + track" or ...
I was trying to express the idea that it was pop industry who made the decision that composers didn't matter rather than pop consumers and maybe that given the information pop listeners might well be happy to know where the pop comes from.
They should (and probably do, behind the scenes) have separate data fields for composer and performer. At least, I believe that labels will input those separately for each track, to enable correct tracking of compensation.
Then Spotify will lump them together into “artist” n the UI, I assume. To make it easier for a casual user to find what they want.
The “performer” Fanny Mendelssohn should probably not be confused with the Composer of the same name, and they should ensure this by using a unique id for each person rather than just a string field.
I like how you think labels correctly label anything. I happen to work at a music streaming service (not Spotify) and I can definitely say that those datasets are insanely noisy and generally you can't really trust labels to, well, label their data correctly for anything.
I was about to say something similar. I don't even work at a streaming service, but I know label-provided metadata is enough of a steaming pile that there are third party start-ups which promise to help with the problem and such.
I don't think it's nefarious, just incompetence. I like Poppy, the bubble gum pop youtube girl turned metal, turned grunge artist. Her name is pretty generic though, and I've seen multiple artists with the same name have their stuff put under her identity on spotify briefly. The artists I've seen are clearly not fake artists trying to sneak into her brand.
Composer / performer is a whole nother metadata thing that should happen.
What if there was a new artist with the same name? Not too crazy. There was another kid in our high school with the exact same first, middle and last name as my brother. No relation.
It happens; what last.fm would do is add the country of origin to the band, like Shining and Shining (nor) to differentiate the Swedish depressive black metal band and the Norwegian extreme avant-garde jazz metal band.
You’d think, she shouldn’t be releasing any new works, you know, on account of being long-deceased and all. Therefore Spotify could block releases claiming her as an artist. Unfortunately, there are new renditions of her work being performed, recorded and published regularly, by many different groups. All of them have the right (perhaps even an ethical duty) to put Fanny Mendelssohn as one of the artists. These are works I’d like to hear, so Spotify’s recommendation algorithms are on the right track.
How on earth can Spotify distinguish real Fanny from fake Fanny?