Do you remember the early World-Wide Web? When this work started there were under a thousand sites altogether, colour images were a novelty, and search didn't exist yet. I was helping teach an intro AI (and Prolog programming) class for Yoav Shoham at Stanford. We set the students an exercise to crawl a little bit of web every day and recommend interesting pages to a user, learning to improve the recommendations over time. It turned out to be an excellent problem. That summer we carried on with Yeogirl Yun, one of the students, and created the LIRA system (Learning Information Retrieval Agents).
It was still an excellent problem, and it became the focus of my PhD. Fab was one of the first recommender systems, running on an ever-expanding network of other students' workstations to recommend web pages to users based on their feedback. Some of the main innovations included combining collaborative and content-based techniques (to cancel out problems of using either approach alone) and giving users control of recommendation accuracy vs. novelty.
It was a very exciting time in the Bay Area. The web was taking off amid a frenzy of entrepreneurship. Marcus Polanco and James Rucker were working on commercial applications of collaborative filtering and recommendation technology with Imana, and I joined them in Potrero Hill to help build a product called CommonQuest.
In the academic world, Recommender Systems has now become a field in its own right, and the early lessons learned from systems like Fab are now foundational elements that continue to be built upon - indeed an early paper we wrote describing Fab now has over 2000 citations.