User demand-driven patent topic classification using machine learning techniques

Traditional patent classification schemes, which are mainly based on either IPC or UPC, are too complicated and general to meet the needs of specific industries. The paper proposes a dynamic classification method, the “user demand-driven patent topic classification”, aiming to a specific industry or technology area. In the paper, classification topics of the method are grouped into technical topic, application topic and application-technical mixed topic. Automatic process of the method using machine learning techniques is presented as well. A case study on the technology area of system on a chip (SoC) is conducted using machine learning techniques, validating the feasibility of the method. The experiment results demonstrate that automatic patent topic classification based on the combination of patents’ metadata and citation information can obtain perfect performance with a greatly simplified document preprocessing.