Text categorization study case: Patents´ application documents

This paper presents computational methods aiming to patent´s text categorization in Portuguese language, involving techniques from machine learning and computational linguistics. The algorithm used was the k-Nearest Neighbor method (k-NN) modified which showed good results, although it requires much computational time in the training stage. For the pre-processing step, it was implemented, with modifications, the stemming method called StemmerPortuguese that includes the removal of suffixes, besides the removal of stopwords and treatment of compound terms.