Unsupervised Learning of True Ranking Estimators using the Belief Function Framework

A variant of the ranking aggregation problem is considered in this work. The goal is to find an approximation of an unknown true ranking given a set of rankings. We devise a solution called Belief Ranking Estimator (BRE), based on the belief function framework that permits to represent beliefs on the correctness of the rankings position as well as uncertainty on the quality of the rankings from the subjective point of view of the expert. The results of a preliminary empirical comparison of BRE against baseline ranking estimators and state-of-the-art methods for ranking aggregation are shown and discussed.