Currently, FanCruft uses a simple “distance based” algorithm to make its recommendations. Suppose that we want to know if you’ll like an anime, called A:
♥A
Well, what we really want is not a distance between that anime and another anime, or an imaginary ideal anime, in an arbitrary “anime-space”, but the probability that you’ll like the show:
P(♥A)
Now, we can probably notice some traits about the anime in question, which are “given.” Here written as TA:
P(♥A|TA)
TA could be any trait, like “has cool villain,” “cosplayable,” or even “fanservicy” (stop snickering out there!) By Bayes’ Rule,
P(♥A|TA) = P(TA|♥A)P(♥A)/P(TA)