A Boosting Algorithm for Label Covering in Multilabel Problems

Proceedings of the Eleventh International Conference on Artificial Intelligence and Statistics |

We describe, analyze and experiment with a boosting algorithm for multilabel categorization problems. Our algorithm includes as special cases previously studied boosting algorithms such as Adaboost.MH. We cast the multilabel problem as multiple binary decision problems, based on a user-defined covering of the set of labels. We prove a lower bound on the progress made by our algorithm on each boosting iteration and demonstrate the merits of our algorithm in experiments with text categorization problems.