A Neural Network Approach for Learning Object Ranking

Leonardo Rigutini, Tiziano Papini, Marco Maggini, Monica Bianchini

Abstract: In this paper, we present a connectionist approach to preference learning. In particular, a neural network is trained to realize a comparison function, expressing the preference between two objects. Such a “comparator” can be subsequently integrated into a general ranking algorithm to provide a total ordering on some collection of objects. We evaluate the accuracy of the proposed approach using the LETOR benchmark, with promising preliminary results.