From a Geekbeat.tv interview (http://youtu.be/4LbyKrtYjM8?t=7m13s), the relevant portion starts at 7:13.
Basically, Yelp uses a mix of:
1. automated filter - computer algorithm,
2. string operations - answer craigslist posts for fake reviewers and then bust the violators, and
3. user reporting - look into cases where users are reporting to the company about possibly fake review.
I see this as an issue with Amazon eBook rating as well, since I look through and download the free kindle books sometimes (via kindlebooklist.blogspot.com). A lot of the eBooks on Amazon's top 100 are of poor quality, and yet these same books often have 5-10 5-star ratings. If I see a book with 5 to 10 5-star reviews and 1 or 2 1-star review, I know there's a very high probability that the book is not worth reading. Of course, one can dig deeper into the issue, such as checking to see if the 5-star reviewers only give 5-star to books and whether the review is overly positive and simplistic. But often, if the book is free and the title seems interesting, I would just get the book and scan it quickly to check the quality.
Presumably, Amazon could have flagged the seemingly obviously fake reviews, but perhaps the company places a very high value in avoiding false positives.