Will AI solve the patent classification problem?

Abstract

This paper scrutinizes the act of patent classification as it is performed by specialists, namely patent examiners, and currently supported by automated systems in patent offices for assigning classification codes to patent application documents. It collectively discusses aspects of the patent classification operation, some of them not very visible, which are not commonly encountered in other document and text classification tasks. The advent of Deep Learning (DL) and, especially, Large Language Models (LLMs) offer a new perspective on the development of automated systems addressing these inherent aspects of patent classification. Towards this direction, the paper analyses how these technologies can address the patent classification problems and concludes with a discussion of potential challenges and benefits that the application of Artificial Intelligence (AI) technologies may bring to the task of patent classification.