Boorah is an alpha site of Palo Alto Web 2.0 startup that extracts information from online reviews to help users find restaurants. Shorter version: an automated Zagat.

“Boorah.com makes it easy to find a great restaurant by presenting a summary of user and professional reviews in a way that has never been done before. Our service offers a unique, searchable database of restaurant locations, cuisines, cost, and hours of operation coupled with the ability to make online reservations. Boorah simplifies the process of dining out whether you’re a “foodie” or just an average Joe looking for a fresh new place to eat.”

If you want to try Boorah you have to get an invitation or signup to request one. I noted a post on Boorah on matthew Hurst’s Data Mining blog and asked for one. Having just gotten it, I was able to explore the service a bit and have a few comments.

Boorah gets its reviews from papers, general review sites like Yelp, topical review sites like restaurants.com, and directly from Boorah users. An important dimension to Boorah’s service is location, and restaurant review sites tend to organize reviews by cities and actual reviews typically give an address — so the geotagging works well here.

Boorah has Zagat style review summaries which consist of a handful of comments selected from reviews. For example:

“Plutos is one of my favorite places to grab a delicious salad.” …. “Also, for 25 cents you can get a slice of fresh baked bread.” …. “The bread that came as a side was fresh.” …. “This is the one place where your salad bar is fresh and there isn’t a chance that someone has sneezed all over your toppings.” …. “Seriously, the sliced steak on top is delish.”

This looks like straightforward opinion mining at the sentence level, though they may be drawing on the last decade’s work on automatic summarization.

“The Boorah summary is made up of quotations excerpted from the full reviews of restaurants. Our summary strives to pull the most relevant, most representative comments from the many reviews we index to give you a quick, informative take on what reviewers like and don’t like about a particular restaurant. Of course, you can also see excerpts of the actual individual reviews, and link directly to the source of each of these reviews.”

One of Boorah’s differentiators is its use of more sophisticated language technology.

“Our system uses patent-pending Natural Language Processing technology to find, summarize and present information from across the web in a way that is far more useful than it’s ever been before.”

This is the most interesting aspect, IMHO. NLP usually implies parsing, or at least part of speech tagging, and mapping the results into some kind of meaning representation. Restaurant reviewing is a great domain for more sophisticated NLP techniques. (Recall Roger Schank’s restaraunt script?). But the domain model for food is really deep. We all know a lot about food, and the kind of people who write reviews know vastly more. A tart sauce on duck is good, but a tart server isn’t. I like coffee with a naturally bitter edge, as long as it isn’t due to being over cooked. It is probably easy to get the low hanging expressions of sentiment, but much knowledge will have to be used to do a really good job. I’d like to learn more about what technologies they are briging to bear on this problem.

All in all, Boorah looks like a promising start.

Related posts:

1. AskIgor – Automated Debugging Service