Abstract Urban recreation and leisure spaces (RLSs) are vital venues for citizens to relax, socialize, and experience urban life. Public perception of RLSs represents a crucial component of urban studies. Traditional Natural Language Processing (NLP)-based methods for analyzing RLS perceptions from social media data face limitations in semantic understanding, multidimensional sentiment differentiation, and spatial recognition. To overcome these limitations, we propose and validate a novel Large Language Model (LLM)-based analytical framework for multidimensional perception assessment of urban recreation and leisure spaces. Through comparative evaluation of three mainstream LLMs (DeepSeek-R1, Kimi, and QWEN), we selected the best-performing Qwen model (F1-score = 0.899) to construct a multidimensional Aspect-Based Sentiment Analysis (ABSA) framework. Taking Hangzhou, China as a case study, we collected note data from the "REDnote" platform and employed the LLM to identify seven perception dimensions and determine sentiment polarities. Building upon the ABSA perception evaluation data generated by the LLM, we further conducted spatial distribution analysis, keyword co-occurrence network analysis, and place-activity bipartite network analysis to reveal perception patterns and their underlying causes. The findings indicate that: a) Hangzhou's recreation and leisure spaces excel in "soft power" dimensions (cultural atmosphere, spatial aesthetics) but require optimization in "hard power" aspects (accessibility, functional configuration); b) Hangzhou's leisure resources are unevenly distributed, with perception evaluations exhibiting a core-periphery structure; c) open outdoor spaces are more positively perceived than enclosed commercial complexes, with "city walking" emerging as the dominant positive activity. This research confirms the transformative value of LLM as core analytical engines rather than preprocessing tools, providing a new research paradigm and practical guidance for the planning and management of urban recreation and leisure spaces.
Bai et al. (Mon,) studied this question.