There are two recommendations systems in Goodreads. One is the official, I agree it is completely broken. The other is "Readers also enjoyed" section on each book page and this one is very good.
It doesn't seem so good and I can't imagine it's using anything sophisticated to make recommendations. What would be great is if it copied Netflix's Cinematch of yore and did SVD on a giant matrix of what you're liked/disliked and what others who have read the same books have liked/disliked.
Isn’t that how literally every one of these is implemented for the last decade?
I don't know, but Cinematch is the last one I remember working well.
its not very good in my experience, within a few "next" clicks you're completely off the path of whatever you were looking for