Your favorite movies are, in order, Freddy Got Fingered, Citizen Kane and The Little Mermaid.
Your favorite TV shows: Washington Week in Review and Championship Bull Riding.
If you could buy just one car, money being no object, it would be either a 2002 Lamborghini Diablo or an ’86 Dodge Caravan.
You probably are one of a kind. But there are others out there who share at least some of your tastes, and if you’re willing to reveal a few of them to John T. Riedl ‘83, he’ll find your soulmates. More importantly from a business standpoint, he’ll help introduce you to products and services your soulmates are buying.
Riedl, an associate professor of computer science and engineering at the University of Minnesota, is the deviser of the algorithms behind software that has become a standard feature of e-commerce sites like CDNow. The software watches the selections you make and suggests items you might also enjoy. By grouping you with others who seem by their purchases to share your tastes, the program can suggest books, CDs or whatever else apparently like-minded people have bought but which you, according to company records, haven’t.
Riedl calls the concept “collaborative filtering” or “personalization-centric enterprise.” Some, including him, see it as a radical new approach to marketing.
“Traditional marketing is guys who want to sell stuff to you by trying to trick you into buying it,” he says. “What we’re into with personalization is doing the best job we can of suggesting the product we have that you might want to buy.”
Riedl says he got the idea for collaborative filtering after he attended a talk in 1992 by a Japanese researcher, Shumpei Kumon, who had spent 10 years studying the evolution of the world economy from prehistoric times to the present. Kumon concluded that the world was moving into an era “when the most important thing wasn’t information but wisdom,” Riedl says. Riedl realized that for computer scientists like him the challenge was to find a way to sift through the ever-widening ocean of facts and figures to find specific wisdom.
Conventional attempts to sort through huge caches of information like Internet web pages have focused on two methods, Riedl says. One is keyword searches, but as anyone who has used a conventional search engine knows, these can return hundreds or thousands of matches “most of which are pages written by high school students doing class projects,” he says.
The other approach is to build a computerized robot to hunt using artificial intelligence. That doesn’t work very efficiently either, he says. “A.I. is the process of having a computer do badly what a human does well.”
He describes collaborative filtering as a system that “lets the humans provide all the quality opinions and the computers do the math.”
In addition to his academic career, Riedl serves as chief scientist and a member of the board of Net Perceptions. Riedl, former Microsoft executive Steven Snyder and others formed Net Perceptions in 1996. The Edina, Minnesota, company markets and licenses collaborative filtering software to clients in the mail-order and e-commerce companies, including Fingerhut, Hudson’s Bay Company and JC Penney.
Eric Meadows, Internet director for Musicians Friend, the world’s largest direct-mail retailer of musical gear, says his company’s website has seen “significant lifts” in sales since it installed Net Perceptions software 2½ years ago. When a customer visits the site and selects a particular electric guitar, for instance, a column on one side of the screen displays other products that purchasers of the same guitar have bought.
Meadows says he can tell that the software has been responsible for the website’s sales gains because sales drop when the software is turned off for system maintenance.
Collaborative filtering may have caught on quickly, but like nearly all technology companies related to the Internet, Net Perceptions has felt the sting of the dotcom bust. According to Riedl, the company’s payroll, which once numbered several hundred, has dropped to less than 100. Its stock, which peaked at $60 a share in January 2000, was trading for less than $2 in March 2002.
Although collaborative filtering has proven a useful marketing device for companies, Riedl envisions a day when it will be used in reverse. Instead of visiting websites or calling mail-order houses and having these companies deduce your tastes, shoppers will carry what he calls a “profile-in-a-pocket,” a private repository of one’s likes and dislikes.
“The time will come when all customer information is contained and maintained by the individual who then determines when, where and why to share that information in exchange for recommendations.”
To judge for yourself how well collaborative filtering can guess your likes and dislikes — without having to buying anything — visit the University of Minnesota’s MovieLens research website, movielens.umn.edu. The site asks you to rate a number of movies and then predicts how well you would like other films.
Ed Cohen is an associate editor of this magazine.