A real customized algorithm based on vector space.
Do you mean representing each user as a vector in an N-dimensional space for N = number of possible votes, then using the distance between them as indication of similarity to weigh their votes relative to your preferences? If so, +1.
Um, yes distance being calculated by a smarter calculation than Pearsons or Euclidean, a variation of RV or something that closely mimicks the way our minds compare for similarity. No bag of words or naive Bayes. True probability.