Classification automatique de courriers électroniques par des méthodes mixtes d'apprentissage

Abstract

New forms of written communication (electronic mail, forum, chat, SMS, etc.) are new challenges for Natural Language Processing methods. These data present very particular linguistic phenomena: too short and very noised messages... This paper focuses on the development of generic tools and resources for e-mails classification. This study deals with the problems of the precise muting of e-mails. After a filtering and lemmatization step, vectorial representation of texts is used for classification purpose by means of supervised, semi-supervised and unsupervised learning techniques. Very good results are presented on realistic corpora.