Group User Models for Personalized Hyperlink Recommendations

摘要

This paper presents a system that combines adaptive hypertext linking based on group link preferences with an implicit navigation-based mechanism for personalized link recommendations. A methodology using three Hebbian-style learning rules changes hyperlink weights according to users' overlapping navigation paths and causes a hypertext system's link structure to converge to a valid group user model. A spreading activation recommendation system generates navigation path based recommendations for individual users. Both systems are linked, thereby combining both personal user interests and established group link preferences. An on-line application for the Los Alamos National Laboratory Research Library is presented.