Facing Uncertainty in Link Recommender Systems
Résumé
Most of prefetching and web recommender systems require a learning phase on a users behavior database. In most of the situations data are the outcome of a preprocessing task of HTTP log files which contain information intrinsically uncertain. This paper deals with modelization of this uncertainty in a link recommendation perspective. A new algorithm based on evidence theory is presented. In addition, new general characterizations of recommender systems are introduced.