IPC multi-label classification based on the field functionality of patent documents

Abstract

The International Patent Classification (IPC) is used for the classification of patents according to their technological area. Research on the IPC automatic classification system has focused on applying various existing machine learning methods rather than considering the data characteristics or the field structure of the patent documents. This paper proposes a new method for IPC automatic classification using two structural fields, the technical field and the background field selected by applying the characteristics of patent documents. The effects of the structural fields of the patent document classification are examined using a multi-label model and 564,793 registered patents of Korea at the IPC subclass level. An 87.2% precision rate is obtained when using titles, abstracts, claims, technical fields and backgrounds. From this sequence, it is verified that the technical field and background field play an important role in improving the precision of IPC multi-label classification at the IPC subclass level.