Measurement and Spatiotemporal Evolution Characteristics of Artificial Intelligence Invention Patents in Guangdong Based on the BERT Large Language Model

Artificial intelligence (AI) has emerged as a key driver of high-quality regional development by reshaping innovation systems, industrial structures, and spatial economic dynamics. Consequently, the scientific measurement of the spatial distribution and evolutionary trajectories of AI technologies has become a critical issue in economic geography. Existing empirical studies typically measure AI activity using enterprise registration data or granted invention patents based on proxy variables, keyword searches, or the International Patent Classification system. However, these methods often suffer from limited semantic accuracy and incomplete coverage, making it difficult to fully capture the rapidly evolving and context-dependent nature of AI technologies. To address these limitations, this study developed a semantic-based identification framework based on large language models. Drawing on approximately 1.2 million granted invention patent abstracts from Guangdong Province between 2001 and 2021, we employed Bidirectional Encoder Representations from Transformers (BERT) large language model to identify AI-related technologies based on deep semantic understanding. This approach yielded a dataset of approximately 200,000 AI-related patents and provided a more comprehensive and accurate representation of regional AI innovation activities. Building on this dataset, we applied BERTopic for topic modeling to identify major technological themes and trace their temporal evolution. The empirical results reveal several key findings. (1) From a temporal perspective, the evolution of AI technologies in Guangdong Province followed a clear two-stage trajectory. During the initial stage from 2001 to 2014, AI patenting activities remained at a relatively low level, gradually increasing from 37 patents in 2001 to 3,514 in 2014. By contrast, the period from 2015 to 2021 represents a phase of rapid expansion, characterized by a sharp increase in AI patenting activities and a substantial acceleration in innovation intensity. This shift indicates the growing strategic importance of AI in regional innovation systems. (2) From a spatial perspective, AI technologies are highly unevenly distributed across Guangdong Province, exhibiting strong agglomeration in the Guangdong-Hong Kong-Macao Greater Bay Area. Shenzhen and Guangzhou together account for 75.1% of all AI patents in the province, forming a pronounced core region of AI innovation. Shenzhen contributed to more than half of the provinces' AI patents, demonstrating a strong primacy position. Beyond these two leading cities, Dongguan, Zhuhai, and Foshan constituted the secondary tier in terms of patent volume. Further analysis of co-invention patents revealed the network characteristics of AI technological collaboration. Within Guangdong Province, inter-city cooperation exhibited a clear dual-core structure centered on Guangzhou and Shenzhen, with dense collaborative linkages concentrated in the Greater Bay Area. While Shenzhen dominates AI patent production, Guangzhou demonstrates the highest level of intraprovincial collaboration, indicating a stronger coordinating and connective role within regional innovation networks. (3) In terms of technological content, topic modeling identified five major AI technology themes: data and image processing, robotics and automation devices, intelligent transportation and fault detection, smart homes and environmental control, and bio-simulation and image analysis. Among these themes, data and image processing constituted the most active and foundational domains throughout the study period, entering a phase of rapid growth around 2013 and peaking in 2019. Robotics, intelligent transportation, and smart home technologies have expanded markedly after 2015, reflecting the increasing diversification and application-oriented nature of AI innovation. By contrast, biosimulation and image analysis exhibited modest growth, suggesting a narrower range of applications. Moreover, cities within Guangdong displayed differentiated thematic advantages, reflecting the distinct trajectories of regional AI specialization. Shenzhen has maintained a leading position in image and data processing, as well as robotics; Guangzhou has developed distinctive strengths in intelligent transportation and urban service applications; Zhuhai integrated AI into its home appliance manufacturing base and marine technologies; Dongguan focused on AI applications in intelligent manufacturing and environmental governance; and Foshan emphasized the integration of smart home technologies with industrial automation.