Attribution d'auteur par ensembles de séparateurs

The authorship attribution problem can be viewed as a categorization problem. To discriminate between different writers (or categories), we must first select a list of useful features (word types in this study), and then we train our classifier. To improve effectiveness, we can consider an ensemble of models instead of a single classifier (bagging). In the current study, we propose two forms of variation: varying the author profiles on the one hand, and on the other, varying the list of selected features. To compare the effectiveness of these solutions, we have extracted two corpora from the Glasgow Herald written by five columnists, the first one is on sports (1,948 articles), and the second on politics (987 articles). Using the KLD model (Zhao & Zobel, 2007), we found that a simple classification scheme tends to produce results comparable to those obtained from using more complex ones. MOTS-CLÉS : Méthode d’ensemble, attribution d'auteur, catégorisation de textes.