Ranking with Unlabeled Data: A First Study
Nicolas Usunier, Vinh Truong, Massih-Réza Amini, Patrick Gallinari, Marie Curie · 2005
In this paper, we present a general learning framework which treats the ranking problem for various Information Retrieval tasks. We extend the training set generalization error bound proposed by [4] to the ranking case and show that the use of unlabeled data can be beneficial for learning a ranking function. We finally discuss open issues regarding the use of the unlabeled data during training a ranking function. 1