Apprentissage semi--supervisé de fonctions d'ordonnancement
Vinh Truong and Massih Amini
In: 7th Conference on "Extraction et Gestion des Connaissances" EGC 2007, 23-27 January 2007, Namur, Belgium.
The growing availability of on--line resources requires the conception of generic approaches that are able to automatically find relevant entities with respect to a user's demand. Recently there has been an increasing interest of the Machine Learning community for the task of ranking by supervised learning of scoring functions. The aim is to learn a mapping from instances to rankings over a finite set of alternatives. Labeling large amounts of data may require expensive human resources, which are unfeasible in most applications. It has been shown in the classification framework that learning with both labeled and unlabeled data may lead to a more efficient decision rule than learning with labeled examples alone. In this paper, we propose a semi--supervised method for the bipartite learning task, which can rank unseen instances. We have led experiments on the real--life dataset CACM gathering titles and abstracts from the journal Communications of the Association for Computer Machinery. The empirical results have shown the potential of this approach in the context of Document Retrieval.