Li Xin, Cheng Haolun, Zhong Xiaofei, Gao Ning
Identifying cutting-edge technologies is of crucial importance for enterprises' R&D strategic decision-making and formulation of governments' strategic plans for scientific and technological innovation. In response to the existing shortcomings in current research on cutting-edge technology identification, namely, the inability of traditional topic models to mine industry-specific technical terms, infer relationships between technological topics and potential application domains, and lack of in-depth analysis on the application scenarios of cutting-edge technology topics, this study proposed a cutting-edge technology identification method based on large language models (LLMs). Taking the intelligent wearables field as an example, the feasibility and effectiveness of this method were validated. The study revealed the following findings: (1) Compared to traditional topic models and clustering methods based on word similarity and word embeddings, LLMs can extract more professional technical information from patent text data and better uncover technical terms and cutting-edge technology topics. (2) The deep semantic parsing capabilities of LLMs can effectively reveal the potential application fields and future application scenarios of cutting-edge technology topics. The systematic correlation analyses of cutting-edge technology topics—potential application fields—future application scenarios provide a foundation for systematically identifying cutting-edge technologies with potential application domains and future application scenarios. (3) The cutting-edge technology identification method based on LLMs, constructed using a ternary coupling analysis approach of "comprehensive indicator evaluation - deep semantic parsing - application scenario analysis", can systematically identify cutting-edge technologies that are not only forward-looking, pioneering, and exploratory but also possess potential application fields and future application scenarios. The cutting-edge technology identification method based on large language models developed in this study has enriched existing approaches and will offers a novel research methodology for the identification of cutting-edge technologies.