The present work proposes a method for the automatic extraction of textual elements within documents, in the specific instance of users within patents. The extraction is made on a massive quantity of patents (almost 40.000) and it is automatic thanks to the integration of text mining and machine learning techniques. The techniques are combined to pursue two conflicting objectives: getting an effective classification tool and, at the same time, obtaining a fast and low-memory algorithm and its classification logics understanding. The management approach to the problem has allowed to keep the machine learning process under control, thanks to the use of the design tool “HP tree”, a hypothesis tree.