Feature Bagging for Author Attribution.
François-Marie Giraud, Thierry Artières · 2012
Abstract The authorship attribution literature demonstrates the difficulty to de-sign classifiers overcoming simple strategies such as linear classifiers operating on a number, most frequent, of lexical features such as character trigrams. We claim this comes, at least partially, from the difficulty to efficiently learn the con-tribution of all features, which leads to either undertraining or overtraining of classifiers. To overcome this difficulty we propose to use bagging techniques that rely on learning classifiers on different random subset of features, then to com-bine their decision by making them vote. 1