When I have the whole body of reviews readily available, I like to just do a Bayesian average. Mix in the average number of reviews at the average review score to keep small data sets from skewing results.