Thursday, October 12, 2006

Restaurant recommendations

The new Zagat's guide to New York Restaurants was released yesterday. Despite the fact that they sent me a free copy in the mail Tuesday (my first blog perk!), I'm not all that thrilled, and it's got me thinking a bit about restaurant recommendations.

There are lots of ways that people hear about restaurants. Sometimes you're just walk or drive past something and go in based on the name. In some markets restaurants advertise on TV or on billboards. Word of mouth is of course a big one. If someone you know mentions a place they like you might trust that as a particularly valuable recommendation. Or then again, you might not. Now, in the age of the Internet, there are new ways of getting reviews. Web sites like citysearch allow ratings and reviews for restaurants as well as hotels and movies and what-not. Menupages takes the ratings and reviews and adds (sometimes out of date) menus for many restaurants as well. Each of these typically have a few to a few dozen ratings and reviews, of people who may or may not have good taste in food, or for that matter your taste in food.

Two more prominent sources for ratings and reviews are Chowhound and the Zagat guide (in paper or online). Chowhound has very informal reviews and comments about all sorts of restaurants, focusing on the out-of-the-way ethnic restaurants and hidden gems. Now that they've updated the software, it's actually possible to search and find reviews of a specific restaurant or ethnicity, but it's organized as a discussion board, not as a database. Zagat, of course, wins in terms of comprehensiveness. According to the press release they sent out (probably to every blogger they could find), the 2007 edition reviews 2,014 restaurants (out of perhaps 10,000 licensed eating establishments in the city), and their iconic 1 to 30 ratings are based on an astonishing 2,100 ratings per restaurant! So there's lots of reason to think that their numbers are meaningful, more meaningful than the average of a dozen ratings on Menupages.

But this is the post-Internet-boom, Web 2.0 age! We expect better now! When we go to, we don't just get a star rating for the books we're looking at, we also get that list of personalized recommendations. Our TiVo knows we like reality TV and automatically records American Idol. Netflix just announced a million dollar prize to whomever can develop an algorithm that beats their current (already rather decent) automatic recommendation system. But for restaurants? We just get aggregate numbers! Pathetic! Old school! Zagat has something like 5 million ratings on three criteria (food, service, decor) by more than 31 thousand reviewers, and they're throwing away their most valuable data, the covariance matrix that allows the individual recommendation systems to work.

I should be able to create an account on, rate a few dozen restaurants, and have Zagat tell me what other restaurants people with my taste (moderate cost, lots of Thai, Indian, French and Italian, higher on food ratings than decor) also liked. It would know that I don't spend a lot of time at expensive seafood restaurants, and wouldn't recommend Le Bernedin. It should take into account my neighborhood, and recommend stuff nearby and for special occasions. It would do a better job of figuring out what I like than I could do. It could find restaurants that me and my three friends, all with accounts on Zagat, will agree on.

Maybe next year.



At 9:36 AM, Anonymous Anonymous said...

Hi Harlan,

You may want to check your sources and your math again. Per a press release "Zagat, New York's biggest-selling restaurant guide, compiled ratings of 2,014 restaurants based on surveys from 31,604 diners this year." 2014 restaurants at an average of 2100 reviews per restaurant, as you claim, would equate to 4.229 million ratings. That would imply that each and every one of those diners rated on average 133 zagat restaurants this year. As someone who is in a related business, it would be as you say, astonishing if it were true, but it's highly likely that it's not.

At 9:49 AM, Blogger Harlan said...

Well, my source is the document that Zagat's sent me. They claim 5.6 million restaurant meals per year, which each of 31,604 surveyers eating out 3.4 times per week, and eating at a reviewed restaurant an average of 11 times. 2100 meals per restaurant is their figure. If I divide the number of ratings by the number of restaurants, I get 2780 meals per restaurant. They must have information about whether there were multiple Zagat ratings at a single table, or something, and adjust for that. 133 restaurants per diner isn't that much -- two new restaurants per week, right? Anyway, although their numbers aren't totally transparent for number-crunchers like us, I am very impressed by them. There are a lot of New Yorkers who eat out most of their meals and rate all of their dining experiences for Zagat.

At 4:50 PM, Blogger Harlan said...

Oh, and folks might be interested in this "scandal" brewing at Menupages:

At 2:28 PM, Blogger Harlan said...

Oops, and of course I forgot to mention newspaper restaurant reviews. The New York Times is of course the most prominent reviewer here, and they have a decent search function. But of course there are only a few reviewers, so their comprehensiveness is not nearly that of Zagat or Menupages or whatever.


Post a Comment

Subscribe to Post Comments [Atom]

<< Home