Automation is necessary when you're talking about finding reviews for over 500K restaurants.

The best solution I can think of right now is 1) collecting valid professional review links from around the web 2) associating them with specific restaurants and 3) performing sentiment analysis to determine a rotten or fresh type rating.

These steps are straightforward and would not yield 100% accuracy. Even a 95% accuracy would be unfair to the few restaurants who's rating somehow gets messed up.

That would be awesome! The main difficulty is crawling for the data, I suppose. Movies are very general, so you have a small set of top critics to pull from, whereas restaurants have reviewers in every city.