GenDesc: A partial generalization of linguistic features for text classification

Abstract

This paper presents an application that belongs to automatic classification of textual data by supervised learning algorithms. The aim is to study how a better textual data representation can improve the quality of classification. Considering that a word meaning depends on its context, we propose to use features that give important information about word contexts. We present a method named GenDesc, which generalizes (with POS tags) the least relevant words for the classification task. © 2013 Springer-Verlag Berlin Heidelberg.