Dynamic models of expert groups to recommend web documents

Abstract

Recently most recommender systems have been developed to recommend items or documents based on user preferences for a particular user, but they have difficulty in deriving user preferences for users who have not rated many documents. In this paper we use dynamic expert groups which are automatically formed to recommend domainspecific documents for unspecified users. The group members have dynamic authority weights depending on their performance of the ranking evaluations. Human evaluations over web pages are very effective to find relevant information in a specific domain. In addition, we have tested several effectiveness measures on rank order to determine if the current top-ranked lists recommended by experts are reliable. We show simulation results to check the possibility of dynamic expert group models for recommender systems.