Perhaps old ratings should be removed or down-weighted in the system as time passes. This would make the ratings better reflect current levels of quality, (a) helping people learn and recover from mistakes and (b) keeping people from coasting on past success while reducing quality.

Not sure how to get general users to rate with 1-star = bad, 3-stars = ok, and 5-stars = great. Sounds like a social engineering problem. It may be more reliable to use trained evaluators, like the system used by Consumer Reports or the Michelin Guide.