Proof of concept of crowd-sourcing "if you liked this, you may like..." functionality on Ao3 works, using tags to try and match content to content. Right now it's got little to no data, so it's going to be hit and miss at best. But the larger the dataset gets, the more accurate it will get.
Stealing your information, or presuming to say how "good" a work is. The info that gets stored is how the tags are ranked (and that's utterly subjective - there's no wrong or right answer there) and a little about the work itself so it can be linked to. That's the lot.
So at its most basic, it tag matches - you tell the search which tags you're most invested in and you'll (eventually) get a list of works those tags. Then it gets a bit more interesting and starts ranking by how those tags were represented.
So, for instance, you're all about Steve/Sam, but a work listing that tag has the main pairing Steve/Tony. For you that work will be lower down the recommended list than a work with the Steve/Sam tag main and centre. There's a bit more than that down the line, but again, no data, no chance.