It's backwards because it's also upside down! Don't worry, you probably are starting to understand logic programming.
The idea is to forget everything you know of what it means to do proof search. Paradigm 1, the familiar paradigm (call it the Proof Construction Paradigm, following Andreoli
as we are all so wont to do) is "I have a set of sequents that I'm trying to make proofs of these sequents because if I can prove the thing I'm trying to prove has a derivation." In the Proof Construction Paradigm, positive polarity = Datalog, negative polarity = Prolog.
The other paradigm, Paradigm 2, is "let's keep a database of all the sequents we have derivations for because if we can grow that database until it includes the thing I'm trying to prove I will, obviously, be done." This is what's called Inverse Method Theorem Proving, and before you say "that sounds insanely inefficient" remember that it's basically what Forward Chaining does but played out at the level of sequents, not atomic propositions - and Sean McLaughlin is I think putting the finishing touches on the Best Theorem Prover For Intuitionistic Logic In The World (TM) right now and it's based on inverse method theorem proving.
It was in the context of Inverse Method that Kaustuv and company realized that polarity switching made Prolog/Datalog happen, but if your computational model is the inverse method, not the Proof Construction Paradigm, you're growing your derivation trees from the sky down instead of the ground up and everything is reversed: negative polarity gives you Datalog (straightforwardly, actually!) and positive polarity gives you Prolog (in kind of a weird way). Kaustuv wrote a summary newsletter article about this upside-down view, I recommend it