Beyond sentiment: Using large language models to decode multidimensional urban park perceptions for enhanced equality

• We systematically benchmarked various deep learning algorithms to interpret perceived park qualities from online reviews. • Our fine-tuned Park-Perception-LLM achieved an accuracy from 83 % to 91 %. • Perception metrics were integrated into an E2SFCA framework to model park accessibility from a quality perspective. • Inequalities in park quality versus quantity, and spatial mismatch patterns are revealed. • Spatiotemporal factors that inform park perceptions are uncovered using multivariate regression. Understanding how people value urban parks is vital for designing liveable and equitable cities, yet the nuanced measurement of park perception proves elusive. While analysis of user-generated content from social media offers scalability over conventional small-sample surveys or objective indicators, deriving effective perceptions beyond basic sentiment, aligned with actionable urban planning variables, remains a crucial challenge. Consequently, a critical gap persists in evaluating inequality in park accessibility through the lens of multi-dimensional quality experiences within diverse urban settings. This hinders truly citizen-centric green space planning and the derivation of insights that transcend quantity. To address this challenge, this study pioneers a systematic benchmarking of large language models (LLMs) and other encoders to classify three perception dimensions—perceived accessibility, usability, and attractiveness—from massive online review data. The fine-tuned Park-Perception-LLM achieved 83–91% accuracy, outperforming other algorithms. Applied to Hong Kong as a representative case of a high-density metropolis, we measured perception scores for 158 urban parks and developed a novel perception-weighted park accessibility metric. This uncovered pronounced inequalities, identifying neighbourhoods with abundant park capacity yet poor experiential quality—a nuanced socio-spatial disparity. Subsequent analysis of perception pathways revealed that diverse facility provision, natural park elements, nearby cultural amenities, and green streetscapes could significantly enhance perceived qualities, whereas monotonous urban forms undermine attractiveness. This LLM-driven framework accurately infers comprehensive perceptions from online reviews, providing a scalable analytical foundation to support planners, designers, and policymakers in integrating perception-based evidence into future park planning, design, and management decisions towards better quality of life.