Divide to Better Classify
Résumé
Medical information is present in various text-based resources such as electronic medical records, biomedical literature, social media, etc. Using all these sources to extract useful information is a real challenge. In this context, the single-label classification of texts is an important task. Recently, in-depth classifiers have shown their ability to achieve very good results. However, their results generally depend on the amount of data used during the training phase. In this article, we propose a new approach to increase text data. We have compared this approach for 5 real data sets with the main approaches in the literature and our proposal outperforms in all configurations.
Origine : Fichiers produits par l'(les) auteur(s)