Learning and Revising User Profiles: The Identification ofInteresting Web Sites

  • Authors:
  • Michael Pazzani;Daniel Billsus

  • Affiliations:
  • Department of Information and Computer Science, University of California, Irvine, Irvine, CA 92697. E-mail: pazzani@ics.uci.edu. dbillsus@ics.uci.edu;Department of Information and Computer Science, University of California, Irvine, Irvine, CA 92697. E-mail: pazzani@ics.uci.edu. dbillsus@ics.uci.edu

  • Venue:
  • Machine Learning - Special issue on multistrategy learning
  • Year:
  • 1997

Quantified Score

Hi-index 0.02

Visualization

Abstract

We discuss algorithms for learning and revising user profiles thatcan determine which World Wide Web sites on a given topic would beinteresting to a user. We describe the use of a naive Bayesianclassifier for this task, and demonstrate that it can incrementallylearn profiles from user feedback on the interestingness of Websites. Furthermore, the Bayesian classifier may easily be extended torevise user provided profiles. In an experimental evaluation wecompare the Bayesian classifier to computationally more intensivealternatives, and show that it performs at least as well as theseapproaches throughout a range of different domains. In addition, weempirically analyze the effects of providing the classifier withbackground knowledge in form of user defined profiles and examine theuse of lexical knowledge for feature selection. We find that bothapproaches can substantially increase the prediction accuracy.