A High-Potential Technology Discovery System Integrating BERTopic and Large Language Models: A Case Study in Hydrogen Fuel Cells

From the dual perspectives of international competition and systems theory, this study leverages patent data to propose and implement a novel system for identifying high-potential technologies by integrating BERTopic and Large Language Models (LLMs). Using the hydrogen fuel cell sector as its empirical domain, the system employs the BERTopic model for in-depth text mining to discern core technical themes. A multi-dimensional quantitative evaluation framework is established, encompassing indicators of strategic value, significance, innovativeness, and high competitive barriers. Subsequently, the Entropy Weight and CRITIC methods are synthesized for indicator weighting, while the TOPSIS comprehensive evaluation method is utilized to rank the potential value of each technical theme. Building upon this foundation, an intelligent classifier is developed with Large Language Models to perform automated thematic categorization and preliminary value assessment of patent documents. The empirical results demonstrate that the system effectively synergizes macro-level technological value assessment with micro-level intelligent patent classification. It thereby furnishes an efficient and reliable intelligent tool for technology R&D navigation, investment decision-making, and competitive landscape analysis, intending to provide a valuable supplement and reference for subsequent research in the field.