GroupLens, presented at the 1994 ACM Conference on Computer Supported Cooperative Work (CSCW) in Chapel Hill, North Carolina, is widely regarded as the paper that introduced automated collaborative filtering. The authors - Paul Resnick, Neophytos Iacovou, Mitesh Suchak, Peter Bergstrom, and John Riedl - built a system to help people cope with the flood of Usenet netnews, the high-volume discussion groups of the early internet.
The core idea was simple and has powered recommender systems ever since: people who agreed with each other in the past will probably agree again. Readers rated the articles they read on a 1-to-5 scale through their news client. Rating servers, which the paper called Better Bit Bureaus, collected those ratings and used the heuristic that a person’s predicted score for an unseen article could be estimated from the ratings given by other readers whose past ratings correlated with theirs. The system then displayed predicted scores so readers could decide what was worth their time.
A deliberate feature of the design was that the architecture was open: news clients and Better Bit Bureaus could be built independently by different developers and still interoperate. GroupLens grew into a long-running research group at the University of Minnesota, led by Riedl and Joseph Konstan, that went on to build the MovieLens movie-recommendation site and release the MovieLens datasets used in thousands of later studies.
Why business readers should care: the neighbor-based prediction method introduced here is the conceptual ancestor of the recommendation engines that now drive a large share of commerce and media consumption. The same “people like you also liked” logic underlies product suggestions, playlists, and social-media feeds.