Service With A Click

New Web technology zeroes in on your tastes to personalize online shopping

I buy a lot of things from catalogs, but very little from the World Wide Web. While shopping the Web is convenient, few retailers' sites match the breadth of their print catalogs. There are no order-takers to tell you if an item is in stock or to suggest alternatives if it isn't. In short, the Web is a less satisfying buying experience than a paper catalog and a toll-free call.

This may be about to change, thanks to a technology recently escaped from the Massachusetts Institute of Technology Media Lab and other computer-science shops. Still known by the geeky name of "collaborative filtering," the idea is simple. By gathering and pooling data volunteered by customers, a computer can determine what products or services people may like. Making this technology work is a challenge, but it has the potential to revolutionize online commerce.

MOVIE THESAURUS. Reel ( is a good place to see the process in action. This site, which doesn't sell anything yet but is in the process of becoming an online video store, uses a couple of different ways to help you find movies you'll like. The Reel Genius asks you to select a film genre, then offers a list of movies of that type. You assign a rating from 1 to 10 for each movie you have seen. The system combines your ratings with the votes of others with seemingly similar tastes--that's the "collaborative" part--to compile a list of suggested films. As you rate the movies the site suggests, the more closely it reflects your tastes. I found that after a while, Reel did a pretty good job of anticipating my interests, except for its insistence that I would like Pulp Fiction.

Reel's Movie Thesaurus allows you to pick a film that you liked, and it will suggest other films you might also find enjoyable. For example, entering Sense and Sensibility returned the obvious suggestions of Howards End, A Room with a View, Emma, and Persuasion. It also offered an offbeat suggestion, Where Angels Fear to Tread.

Books pose a much greater challenge for filtering than movies. That's partly because there are so many more titles, partly because you only have authors, not actors, directors, producers, and screenwriters, to serve as points of comparison. But, the big Web bookstore ( is trying. Using a technology called GroupLens, developed at MIT and the University of Minnesota and commercialized by startup Net Perceptions, Amazon is developing a system similar to Reel's for rating and suggesting books. An experimental version I tried was a bit flaky--it kept suggesting Danielle Steele novels. The developers say there's currently little preference information in the database, but it should improve greatly before the system goes public later this spring.

To make this guided shopping work, Web sites have to collect a great deal of personal information, which marketers would pay dearly for. This naturally raises concerns about privacy that Web merchants have had to address. For example, FireFly Network, ( a Media Lab spin-off that uses filtering to recommend entertainment and to create online communities of like-minded people, has a formal privacy policy that promises not to turn personal information over to third parties without the express consent of the individual.

That's probably a necessary step to win consumer confidence in Web commerce based on guided shopping. Analysis that more quickly homes in on consumers' tastes is also needed. I doubt that collaborative filtering will ever equal the instincts of a good Nordstrom salesperson. But as the technology moves out of its infancy, online shopping could challenge those trips to the mall.