Dynamic expert group models for recommender systems

Abstract

Recently many recommender systems have been developed to recommend items in online commerce markets, 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-group models to recommend domain-specific items or documents for unspecified users, while users give feedbacks of relative ratings over the recommended items or documents. In this system, the group members have dynamic authority weights depending on their performance of the ranking evaluations. We have tested two effectiveness measures on rank order to determine if the current top-ranked lists recommended by experts are reliable.