Quantitative Process of Retrieving Documents on Specific Technology Using an Academic Database and the Sentence Transformer

ABSTRACT As competition in technologies such as semiconductors and AI intensifies, the importance of technology analysis in establishing R&D strategies is increasing. Although frequently used, technological document searches by domain experts often suffer from limitations in qualitative judgment and inconsistency. This study proposes a process for quantitatively assembling a collection of documents on specific technologies using an academic database and a language model. Using approximately 92 million records from the Web of Science, we identified related keywords from author keywords and used scientometric approaches to determine a core set of technology‐related literature. Then, a deep learning‐based Sentence Transformer model was employed to extract technology literature with high similarity to the documents, ultimately forming a set of documents for technology analysis. This study aims to overcome cognitive limitations and introduce quantitative criteria for technology analysis, which is particularly significant for data‐driven science and technology policy.