Briefs

"Making Intelligence a Bit Less Artificial" New York Times (05/01/03) P. E1; Guernsey, Lisa

Amazon, NetFlix, and other online retail services rely on automated recommender systems to anticipate customer purchases based on past choices; however, a February report from Forrester Research found that just 7.4 percent of online consumers often bought products recommended by such systems, roughly 22 percent ascribed value to those recommendations, and about 42 percent were not interested in the recommended products.

To improve the results requires the enhancement of recommendation engines with human intervention, according to TripleHop Technologies President Matt Turck. One of the key ingredients of today's recommendation technology is collaborative filtering, in which a buyer is matched to others who have bought or highly rated similar items. Commonplace problems with this methodology include cold starts, in which predicting purchases is difficult because the system lacks a large database of people with similar tastes, and the popularity effect, whereby the computer delivers recommendations that are pedestrian and prosaic.

Some companies try to avoid such problems by adding a human element: Barnesandnoble.com, for instance, employs an editorial staff to tweak recommendations. "If it is not vetted and monitored by humans and not complemented by actual hand-selling, as we say in the book industry, it doesn't feel like there is anybody there," notes Barnesandnoble.com's Daniel Blackman.

Here is the full story.