Automatic users extraction from patents

Patents contain a large quantity of information which is usually neglected. This information is hidden beneath technical and juridical jargon and therefore so many potential readers cannot take advantage of it. State of the art natural language processing tools and in particular named entity recognition tools, could be used to detect valuable concepts in patent documents. The purpose of the present research is to design a method capable of automatically detecting and extracting one of the multiple entities hidden in patents: the users of the invention. The method is based on a new approach tailored for users extraction by integrating state-of-the-art computational linguistics tools with a large knowledge base. Furthermore the paper shows a comparison among different machine learning algorithms with the twofold aim of achieving the highest recall and evaluating the performance in terms of precision and computational effort. Finally, a case study on two patent sets has been conducted to evaluate the effectiveness and the output of the entire tool-chain.