An Extension of RankBoost for Semi-Supervised Learning of Ranking Functions
The purpose of this paper was a semi-supervised learning method of alternatives ranking functions. This method extends the supervised RankBoost algorithm to combines labeled and unlabeled data. RankBoost is a supervised boosting algorithm adapted to the ranking of instances. Previous work on ranking algorithms has focused on supervised learning (i.e., only labeled data is available for training) or semi-supervised learning of instances. The authors are interested in semi-supervised learning, which has as objective to learn in the presence of a small quantity of labeled data, simultaneously a great quantity of unlabeled data, to generate a ranking method of alternatives.