Using IR techniques to improve Automated Text Classification

This paper performs a study on the pre-processing phase of the automated text classification problem. We use the linear Support Vector Machine paradigm applied to datasets written in the English and the European Portuguese languages – the Reuters and the Portuguese Attorney General’s Office datasets, respectively. The study can be seen as a search, for the best document representa- tion, in three different axes: the feature reduction (using linguistic in- formation), the feature selection (using word frequencies) and the term weighting (using information retrieval measures)