Welp, sounds like Yelp's algorithm for this is sort of wacky. Ratings supported by fewer data points (which I would call less reliable for that reason) are over-represented.
Seriously. I worked in research for a while and we would weight data to fit demographics but there were limits to how much weighting you could apply before the data sample was considered insufficient.
I really wonder how much this algorithm is ignoring general sampling rules like ‘n > 30’ to crank out this list.