On NPR recently, there was a guy talking about a startup that works like Tripadvisor, but only shows you reviews within your social circle. This eliminates shills and only shows you opinions that have a known value to you.
Maybe someone can steal that idea and apply it to restaurants.
I think there's a golden opportunity that's less about "social circles" and more about "revealed common preferences".
If there's somebody on your review site, who ranks highly several of my highly ranked coffee shops, and ranks low a subset of my lowly ranked coffee shops, I'd be much more prepared to consider their other coffee shop reviews as relevant to me. Even better, location based services like 4Square of Facebook checkins could expose their "revealed preferences" to the algorithm as well, someone who rates Fourbarrel and Ritual highly, but checks in to Starbucks four times a day is less likely to be someone who's reviews I'd want to read than someone who perhaps rates Fourbarrel and Ritual lower on an absolute scale, but checks into both regularly as well as, say Sightglass and BlueBottle.
I'd really like a time and event aware as well as just venue aware review function. Zeitgeist on a Thursday night, or after a critical mass ride - is a _vastly_ different experience than Zeitgeist on a sunny Saturday afternoon when it's full of slumming sunset and marina crowds. DNA Lounge really needs separate review categories for Death Guild nights, Bootie nights, and out-of-town dubstep artist nights - people who love (and hence rate highly) one of those events are significantly less likely to enjoy the other two as much - which makes a Yelp-style single rating for DNA Lounge not particularly useful.
There is a Swedish movie review site (http://filmtipset.se) which gives you a predicted rating determined from your previous reviews. They are venturing into books and wine, using the same algorithm. Unfortunately they seem to be better mathematicians than web developers, so all of their sites are pretty crappy and they seem to have problems with monetization.
Another problem is that if your taste is not really mainstream, you have to rate quite a few movies to get accurate predictions. After I reached a couple of hundred movies, the predictions were nearly always completely accurate.
We're actually working on this from a slightly different angle; rather than "revealed preferences" and machine learning, we're using surveys of psychologically-validated, taste-predicting traits.
Tripadvisor et al have the advantage that, with all the Internet posting (admittedly with dubious quality, but...), they can show a review of almost anything.
Within my social circle? They won't even be 10% comprehensive for my own town, let alone once I start going away.
What if they expand the data by including reviews from people of second and third order of separation from the user? Each iteration would exponentially expand the data, and perhaps after four or five separations, you'd have sufficient data (though this may defeat the whole premise of getting data from your social circles)
But the further out you go, the lower the quality of the personal link.
What counts as a personal link, anyway? There are people on my Facebook profile I knew years ago in a different town, people on my LinkedIn profile I worked with 6-7 years ago. All people with whom I'm happy to stay in contact at a low level and retain in my wider professional network, but would I want to trust a recommendation from a similarly distant contact of theirs as much as I would a day-to-day colleague of my brother-in-law who I see every week? Probably not.
Maybe someone can steal that idea and apply it to restaurants.