As effective nature-based solutions, urban parks play a critical role in design practice and environmental management, making it essential to understand how their landscape characteristics and surrounding environmental attributes influence public preferences. This study employed a large language model to conduct fine-grained, aspect-level preference analysis, revealing the nonlinear relationships between park and environmental features and public preferences, and highlighting the substantial potential of large language models for related research. The results indicate that park area and total edge exhibit positive threshold effects, with critical values of 22 ha and 30,000 m, respectively. Ample and well-connected blue-green spaces, high-quality vegetation cover, moderate morphological complexity (with an LSI threshold of 30) and appropriate facility provision, together with effective control of crowding and excessive hardscape development, play a crucial role in enhancing public preferences for and experiential quality of urban parks.