Perception and interregional flows of cultural ecosystem services in a megacity: Integrating large language models and the metacoupling framework

Cultural ecosystem services (CES) are fundamental to human well-being, yet their inherent subjectivity and non-material nature pose significant challenges for large-scale quantification and the measurement of interregional flows. To address these challenges, this study investigated the spatiotemporal perceptual patterns of CES and interregional cultural ecosystem service flows (CESF) in Shanghai by developing a definition-driven zero-shot classification workflow based on large language models (LLMs) using user-generated content from Ctrip. Review data from 153 points of interest in 2019 and 2023 were comparatively analyzed to examine intra-urban CES dynamics, and user IP location data were further integrated within a metacoupling framework to characterize the coupling structure and spatial spillover patterns of cross-provincial CESF. Results indicated that the LLM-CES approach achieved high accuracy and stability in multi-category identification, providing a scalable pathway for extracting CES perceptions from large-scale social media data. From 2019 to 2023, CES in Shanghai exhibited a pattern of quantitative decline, structural stability, and spatial contraction. At the national scale, CESF demonstrated a distance-decay pattern characterized by high-intensity short-distance interactions and broad long-distance diffusion. This study advances CES research by integrating perception and flow within a unified analytical framework and offers a transferable methodological approach for assessing megacity CES and their interregional impacts, with implications for well-being-oriented urban planning and cross-regional governance.