Dynamic Identification of Emerging Cigarette Processing Technologies: Integrating Large Language Models and Patent Maps

Cigarette manufacturing technologies are evolving rapidly, making the timely identification of emerging technological directions increasinglychallenging.Conventionalapproachesoftenrelyonkeywordmatching,shallowsemanticanalysis,andextensiveman ual annotation, limiting their ability to capture complex technical concepts and dynamic patterns of technological change. This study develops a dynamic framework that integrates large language models (LLMs) with patent mapping for the identification of emerging technologies in cigarette manufacturing. The framework is evaluated using 9,006 Chinese patent records published be tween2014and2025.Fewshotpromptingandasystematicpostprocessingprocedureareusedtoextractandstandardizetechnical elements from patent texts. Generative Topographic Mapping (GTM) is then employed to trace the evolution of technology topics across multiple time windows. Candidate topics are further evaluated in terms of novelty, technological attention, relative growth, and influence, with indicator weights determined using a goalprogrammingbased BestWorst Method (BWM). The framework enables semantic information extracted from patent texts to be jointly analyzed with temporal patterns of technological evolu tion while reducing reliance on manually annotated data. The results indicate that cigarette primaryprocessing technologies have gradually evolved from a predominantly equipmentcentered structure toward a dualcore configuration combining conventional process optimization with intelligent sensing, control, and digital technologies. Two distinct patterns of emerging technologies are identified. Rapidly emerging topics, represented by imagebased online defect inspection, show strong novelty and growth, whereas persistent topics, including tobaccodrying optimization and adaptive compensation in highspeed cutting, remain active throughout the observation period and constitute important technological capabilities in cigarette manufacturing. The proposed framework provides a systematic approach for combining semantic extraction, dynamic patent mapping, and multicriteria evalu ation to identify emerging technologies and trace their evolution, thereby supporting technology intelligence and R&D planning in cigarette manufacturing.