Extracting user interests from bookmarks on the web

Abstract

This paper regards bookmarking as the most important information to extract user preferences among user behaviors. Bookmarks are categorized on Bayesian networks by an ontology. Considering the relationships between categories, evidential supports are mutually propagated to improve the coverage of the potential preferences. Consequently, we have attempted to define bookmarking behaviors and apply them to the weight updating on users' preference map. We have measured the causal rate in order to improve accuracy of evidential supports and retrieved relational information between the behavioral patterns and user preferences throught temporally analyzing these patterns. For experiments, we made a dataset organized as 2718 bookmarks and had monitored 12 users' behaviors for 30 days(1).