Abstract (EN)

Given a large set of items and a set of users, we consider the problem of collecting user preferences – or ratings – on items. The paper describes a simple method which provides an approximate solution to the problemwithout requiring each user to rate each item. The method relies on an iterative process. Each step, or ballot, requires each user to rate a sample of the items. A collaborative filtering algorithm is then performed to predict the missing ratings as well as their level of confidence (which is initially 0). Perfoming a new ballot allows to improve the accuracy of predictions. The administrator of the system is responsible for stopping the iteration when a satisfactory level is reached. We apply this method to the assignment of reviewers to papers prior to the review phase of conference management, and describe its implementation in the MYREVIEW web-based system.