I really like Yelp, which is probably why I’ve bothered to spend time typing up reviews for it, despite it being a commercial service that could theoretically pull a CDDB at any time. I’ve found a lot of neat little restaurants that I wouldn’t have otherwise found, particularly while traveling, via Yelp, and in general have found the ratings and reviews there to be of very high quality.

However, I’ve noticed that as Yelp’s userbase has grown and expanded beyond the computer-savvy foodie demographic that seemed to have been some of its first users, the average ratings for a particular business are no longer as useful as they once were. It used to be, if a restaurant had five stars and more than a handful of ratings, it was almost certainly phenomenal. Similarly, if a place was languishing at one or two stars, it was probably best avoided — after all, if a place is bad enough to actually get someone (who isn’t being paid) to spend the time to write a negative review, something must be pretty wrong. And if something was in the middle, chances are it was pretty much just average for whatever cuisine it was trying to represent.

Lately, though, I’ve noticed that many places — and this is especially true of eclectic or “acquired taste” restaurants — are getting pushed towards middling reviews not because anyone is actually rating them that way, but because very good and very bad reviews are being averaged out into two or three stars. This isn’t really surprising: reviewing restaurants is a “matter of taste” practically by definition. But that doesn’t make the result very useful. When I’m looking down the search results in Yelp, I want to know what I am likely to enjoy, not what some hypothetical “average user” is going to like. (I’m not the first to notice this problem, either.)

As more and more users join Yelp and start writing reviews, the average review will naturally start to approach what you’d get from reading the AAA guide, or any other travel or dining guide aimed at a general audience. That’s not necessarily bad, and when you’re writing a travel book or dining guide it’s pretty much exactly what you want: try to give an idea of what most people will think of a particular restaurant.

But that’s certainly not the best that an interactive system can do, not by a long shot. The benefit of a website, as opposed to a book, is that the website doesn’t necessarily have to show the same exact thing to everyone. This is why the front page of Netflix is more useful than the top-ten display down at your local Blockbuster, or why Amazon’s recommendations are typically more interesting than whatever happens to be on the aisle-end display at Borders. It’s not that Blockbuster or Borders aren’t trying — they’re doing the best they can to please everyone. The beauty of a dynamic website is that you don’t have to try to please everyone with the same content; you can produce the content in a way that’s maximally useful to each user.

If Yelp took this approach, ratings from users who tend to like the same things that I do would be weighted more heavily when computing an establishment’s overall score; if you brough up the same restaurant (or if it came up in your search results, more importantly), it might have a different score, if your preferences — as expressed via your reviews — are significantly different than mine. This makes perfect sense, and provided that there’s still some way to see what the overall, unweighted average review was (Netflix shows it in small print below the weighted-average), it’s a no-lose situation from the user’s perspective.

I’m sure that Yelp’s engineers are aware of the Netflix model and how it could be applied to Yelp, so this isn’t a suggestion so much as a hope that it’ll get implemented someday.