Greek patent classification using deep learning

Abstract

Patents are documents that contain state of the art technical and scientific information in almost every field of science. Patent applications filed every day in the Hellenic Industrial Property Organization, have to be intellectually classified by domain experts based on a hierarchical taxonomy. As a result, this classification process is labor intensive and overwhelming for the experts. This paper proposes the use of Natural Language Processing (NLP) for automated pa-tent classification. We compare state of the art deep learning-based NLP methods for the automated semantic categorization of these documents. For this purpose, we built a dataset comprised by around 70,000 Greek patents applications, used to train several deep learning algorithms with Greek-BERT obtaining the best accuracy scores.