I decided on a logarithmic weighted bayesian ranking. I read about the typical bayesian ranking (where you have some sort of minimal required number of reviews), then modified the it to work with power-curve distributed sets of data (as #reviews tends to be). Works great and has a few useful levers that I can adjust as I see fit.
Have you by any chance taken a look at the Wilson score? It's a modification of the typical Bayesian ranking that functions much better with small samples.
