When I was looking into writing recommendation systems, one paper made the interesting observation: in a 1-10 rating system, the only places where people will naturally agree upon is 1,5 and 10. Anything else can only be evaluated relative to the same user's other scores. (Comparing my 9 and 10 scores has meaning, but not comparing your 8 and my 9)
The polarization is a natural result of the lack of definition
