Collaborative filtering attempts to alleviate information
overload by offering recommendations on whether information is valuable based
on the opinions of those who have already evaluated it. Usenet news is an
information source whose value is being severely diminished by the volume of
low-quality and uninteresting information posted in its newsgroups. The GroupLens
system applies collaborative filtering to Usenet news to demonstrate how we can
restore the value of Usenet news by sharing our judgments of articles, with our
identities protected by pseudonyms.

This paper extends the original GroupLens work by reporting
a significantly enhanced system and the results of a seven week trial with 250 users
and over 20,000 news articles. GroupLens has an open and flexible architecture
that allows easy integration of new newsreader clients and ratings bureaus. We
show ratings and prediction profiles for three newsgroups, and assess the
accuracy of the predictions.