1PhD Candidate, Department of Industrial Engineering, Konkuk University, Republic of Korea
2Director, AI Lab, Neopons Inc., Republic of Korea
3Master’s Student, Department of Industrial Engineering, Konkuk University, Republic of Korea
4Principal Researcher, Future Technology Analysis Center, Korea Institute of Science and Technology Information, Republic of Korea
5Director, Future Technology Analysis Center, Korea Institute of Science and Technology Information, Republic of Korea
6Professor, Department of Industrial Engineering, Konkuk University, Republic of Korea
Correspondence to Janghyeok Yoon, E-mail: janghyoon@konkuk.ac.kr
Volume 19, Number 3, Pages 155-180, September 2024.
Journal of Intellectual Property 2024;19(3):155-180. https://doi.org/10.34122/jip.2024.19.3.8
Received on July 02, 2024, Revised on August 06, 2024, Accepted on September 03, 2024, Published on September 30, 2024.
Copyright © 2024 Korea Institute of Intellectual Property.
This is an Open Access article distributed under the terms of the Creative Commons Attribution-NonCommercial-NoDerivatives (https://creativecommons.org/licenses/by-nc-nd/4.0/) which permits use, distribution and reproduction in any medium, provided that the article is properly cited, the use is non-commercial and no modifications or adaptations are made.
Patents, i.e., the output of research and development (R&D) activities, are regarded as a concentration of Problem–Solution information. Despite various patent analysis studies aimed at solving problems, large language model (LLM)-based studies are scarce. LLMs, which are effective for natural language processing tasks, such as text summarization and generation, have been applied in numerous fields, including healthcare, finance, and law. By learning the Problem-Solution information of patents as an LLM instead of merely examining existing R&D solutions, one can generate new solutions applicable to a specified problem. Therefore, this study proposes an approach to generate and analyze new R&D solutions using LLMs. Our systematic approach involves 1) collecting numerous patents and constructing a database; 2) extracting Problem-Solution information from the Common Application Form section of patents and constructing a Problem-Solution dataset; 3) fine-tuning an LLM using the problem-solution dataset and generating R&D solutions; and 4) analyzing R&D solutions to present a technology concept portfolio map. This study extends beyond the existing R&D solution exploration, presents a new approach for generating solutions, and suggests technology concepts using LLMs. Therefore, this study contributes to the expansion of the available options and fosters innovation in R&D field.
Patent analysis, Problem-Solution information, R&D solution, Large language model, Fine-tuning
No potential conflict of interest relevant to this article was reported.
The author received manuscript fees for this article from Korea Institute of Intellectual Property.