Improving the Performance of a Korean Patent Document Search Model using KorPatBERT-based CPC Classification Model

Abstract

The global competition for technological supremacy is intensifying, prompting every country to focus on securing technological advantages through patent acquisition. In this environment, efficient and accurate patent searching is a key factor for establishing national technological sovereignty and strengthening global competitiveness. However, identifying prior art patents accurately and effectively within vast patent data remains a challenging task. To address this challenge, this study proposes an advanced patent search model that leverages artificial intelligence technology. This study presents a method for creating models according to the CPC classification model based on the KorPatBERT(Korean Patent BERT) that can deeply understand the detailed technical context of patent documents through pre-training involving vast patent data. Furthermore, this study presents a method for generating high-dimensional document embedding vectors that can effectively reflect the technical subject and context of patent documents and a method for building a search system capable of processing large volumes of patent data in real time. By integrating the proposed patent search model into this system, the study successfully demonstrated improved search performance compared with existing methods in objective performance evaluations. This study can contribute toward enhancing industrial applicability and practical usability by applying the processes of currently operational patent search data and systems. The current study’s findings are expected to provide a foundation for nations and companies to continuously lead innovation and efficiently manage and utilize patents.

Keywords

• intellectual property rights
• patent
• korpatbert
• artificial intelligence
• prior-art patent
• cpc
• patent classification
• patent search
• embedding vector