In the age of machine learning, I'm really surprised there aren't superhuman music recommendation algorithms. Or maybe there are, and these algorithms simply don't serve the corporate interests. But then where are the open-source alternatives?
Because music is extremely hard to quantify. What do you quantify it on? See https://everynoise.com/ (the mess on the page is quantifying by just three or four out of 17 IIRC parameters) and their small doc on it: https://everynoise.com/EverynoiseIntro.pdf
And doing that at scale across hundreds of millions of users quickly becomes prohibitively expensive. So companies simplify, and reach for simpler solutions, unfortunately.
I enjoy using last.fm, although it's not their focus these days. Sign up, connect it to Spotify or whatever you use (incl. a long list of players of local music), after a day or so it'll learn what you like and you can create playlists with suggestions and export them, or browser around recommended artists etc.
I liked Tidal's recommendations.
I went back to last.fm, music stores, friends recommendations, and music/TV scores(a lot of good movie sound folks are amazing musicians).