This paper presents an empirical study of machine learn techniques to text categorization. Specifically aim to classify news about coffee market according with categories from coffee supply chain. The objective is to measure the performance of three types of algorithms: Naïve Bayes based, Tree bases and Support Vector Machine (SVM). A database with news collected from web and labeled by human expert analysts is used in a learning phase. Then automatic classify news extracted from web following the same steps and terms as human according to their relevance for each learned category. The test in a real database shows a better performance by Naïve Bayes based Algorithms for this specific case.