Comment by sthuck
Comment by sthuck 2 days ago
The best ”algorithm” for discovering new music was digging through profiles on last.fm back when the social functions of the site were still active. Sure, it was a lot of manual work, but the results were amazing. It wasn't completely blind, I found that people I had high similarity with, it was more likely I'll like what they like, even across different genres. Sometimes people were nice and took the effort to recommend based on my profile. I got introduced to varied music, different genres and even a bit from different countries.
The worst was Pandora, which did recommendations based on breakdown of musical instruments and elements in the song. It did what it aimed to do pretty well, only it was a bad idea. It gave you a lot of uninspiring music that sounded like a bland copy of something you actually liked.
Spotify's recommendations are not super awful, but definitely feel closer to Pandora's style. I wonder why is the result like that even though I'm sure they train their model based on listening history.
I used to pay for their radio service, it was a bit like Pandora. I found it when they added it to Xbox 360 as an app.
I really liked their original profile pages that had sort of a MySpace style customization & vibe. You could have your favorite musicians and tracks analyzed through their API by these 3rd party services that would create very cool graphics & charts to show off to friends and visitors what you were into.
But, then I guess they ran out of money and were really trying to get scooped up by Spotify. They turned off their music player, disabled all the profile customization, alternative services quit having built in scrobbling to it.
I remember I had to download an app that would constantly have my microphone open and it would ID the song I was listening to via some kind of Shazam service and send it to last.fm. I never considered what a security risk that was because I was more interested in keeping my last.fm music tracked.