Understanding public perceptions of cultural ecosystem services in urban coastal wetland ecological restoration areas: A social media-based large language model approach.

Understanding public perceptions of cultural ecosystem services (CES) in urban coastal wetland ecological restoration areas was essential for coastal resource management and sustainable development. Although social media data has been increasingly utilized to develop CES indicators, significant technical challenges remained in conducting CES assessments and analyzing the primary influencing factors intelligently, accurately, and efficiently from large volumes of textual comments. To address these challenges, this study developed two artificial intelligence (AI)-based methods-using large language models with prompt engineering-to automatically identify CES categories and associated sentiments, and to analyze the key influencing factors. Using the coastal wetland ecological restoration area in Xiamen, China, as a case study, the results indicated that recreation (31.69%) and aesthetics (23.54%) were the two most commonly perceived CES categories. The average sentiment score across the nine CES categories in all restoration areas was positive (4.0-4.6). Differences in CES perceptions among the three distinct types of restoration areas-mangroves, beaches, and bays-were minimal. Public perceptions were primarily influenced by the ecological environment, historical culture, and management practices. These findings provide targeted recommendations for improving restoration planning and sustainable management in urban coastal wetlands. This study demonstrated an innovative interdisciplinary integration of computer science and marine ecology, highlighting the advantages of AI in advancing CES research and offering a new paradigm for understanding public perception.