Using frequency-of-mention in public conversations for social filtering

Abstract

We report on an investigation of using Usenet newsgroups for social filtering of Web resources. Our main empirical results are: (1) for the period of May ’96 to Jul ’96, about 23% of Usenet news messages mention Web resources, (2) 19% of resource mentions are recommendations (as opposed, e.g., to home pages), (3) we can’automatically recognize recommendations with at least 90% accuracy, and (4) in some newsgroups, certain resources are mentioned significantly more frequently than others and thus appear to play a central role for that community. We have created a Web site that summarizes the most frequently and recently mentioned Web resources for 1400 newsgroups.