In this paper authors study the effect of recommender systems (CF) on sale diversity. Motivated by "Lorenz curve", They use Gini coefficient (G) for measuring the bias of a recommender system. First, they used the so-called "urn-model" to explore the biases analytically. In this setting a stochastic function ( e.g., sigmoid function) gives the probability that an item being recommended by the system at each point of time, based on the current marker share. THE MODEL SUGGESTS the average increase of concentration bias for different settings of the model.
S. McNee, J. Riedl, and J. Konstan. CHI '06: CHI '06 extended abstracts on Human factors in computing systems, page 1097--1101. New York, NY, USA, ACM, (2006)