Candidate Ranking method

Hi, could you please explain how is done the candidate ranking? I see that the classifier is fed with (positive - negative) or (negative - positive) candidates vector, but then how is it done in the resolving part?

Thanks a lot.