This is an important topic. I'd like to learn about other, more stochastic methods to help recommenders break free from the simplistic results we currently get. Aren't there ways to model a surface to show multiple peaks and valleys of interest, instead of just one?