Datasets

The preliminary experiments were carried out by using three different corpora: the R8 version of the Reuters collection (train and test) and, partially, a reduced version of the 20 Newsgroups named ``Mini20Newsgroups''. We have pre-processed each corpus eliminating punctuation symbols, stopwords and, thereafter, applying the Porter stemmer. The characteristics of each corpus after the pre-processing are given in Table 1.

Table 1: Characteristics of Reuters-R8 and Mini20Newsgroups

	R8-Train	R8-Test	Mini20Newsgroups
Size	$\approx$2,500 KBytes	$\approx$900 KBytes	$\approx$1,900 KBytes
Documents	5,839	2,319	2,000
Categories	8	8	20