抄録
Amid the rapid development of digital geography, public cognition of urban space increasingly exhibits subjectivity, diversity, and dynamic evolution, while traditional geographical approaches face limitations in uncovering the semantic complexity and spatial heterogeneity of this “perceived city” phenomenon. Drawing on user-generated text data from the Weibo platform in Chengdu, this study constructs a human geography research framework integrating Large Language Models (LLMs) to systematically examine the semantic expression and spatial distribution of urban spatial cognition. Through semantic extraction and sentiment analysis, combined with geocoding and spatial clustering methods, it identifies differences and patterns in spatial cognition among various user groups (e.g., residents versus tourists). The results reveal significant “cognitive hotspots” and “semantic blind zones” in certain areas of Chengdu’s social media landscape, reflecting the misalignment between subjective perception and objective urban structure and highlighting the symbolic reconstruction of urban space in the digital context. As the first empirical study to systematically introduce LLMs into Chinese urban perception geography, this work expands methodological pathways for urban cognition research and, at the practical level, provides theoretical and data support for smart city development, city branding, and spatial perception equity, thereby promoting the cross-disciplinary integration and paradigm innovation of artificial intelligence and human geography.
