Entity resolution is a thing though, and the Discover Weekly algorithm actually tries to combine artists (possibly due to collaborations), which is what goes wrong.
For example, I follow Cornelius, the Japanese guitarist. I liked several of his song. So the data indicates I’m a fan. I’ve been recommended a new song by a different Cornelius. This other guy is an African gospel singer. I don’t follow him. Never even played a song.
That shouldn’t happen. The system knows they’re different Corneliuses, because the catalogs aren’t mixed, but yet string match took over.
I wonder if, instead of a string match, enough people have searched “Cornelius” and (accidentally, out of curiosity, etc.) played the gospel singer’s songs that it’s algorithmically linked the gospel singer to the other artists these people listen to, which are then linked back to the Japanese guitarist.
For example, I follow Cornelius, the Japanese guitarist. I liked several of his song. So the data indicates I’m a fan. I’ve been recommended a new song by a different Cornelius. This other guy is an African gospel singer. I don’t follow him. Never even played a song.
That shouldn’t happen. The system knows they’re different Corneliuses, because the catalogs aren’t mixed, but yet string match took over.