Abstract

Patent analysis offers a useful tool for monitoring technological changes by transforming large volumes of patent data into valuable technological intelligence. However, current approaches are constrained by the large scale of patent datasets and the complexity of documents, requiring skilled analysts and a heavy reliance on domain experts to ensure accuracy. To support more agile technology planning, it is essential to reduce analysts' workloads and mitigate dependency on domain experts. This study investigates the applicability of Large Language Models (LLMs) in patent analysis by evaluating their performance in tasks traditionally handled by experts or analysts. Specifically, the accuracy of LLM-based patent analysis is assessed through a quantitative evaluation using F1 and ARI scores, and qualitative evaluation by domain experts. As a result, this study explores the possible use of LLMs to core patent analysis tasks including data collection, preprocessing, analysis and representation. The findings indicate that although LLMs have significant potential to enhance efficiency in the patent analysis process, further validation and continuous performance improvements are required, particularly in incorporating emerging trends and filtering out irrelevant content from patent search strategies.