Analysis of spatial quality factors affecting underground space development efficiency driven via multimodal data and large language models

To overcome the limitations of traditional evaluation methods in quantifying users' subjective perceptions of underground space design and their impacts on spatial efficiency, this research takes 188 typical underground public spaces in Shanghai as research samples. Based on multimodal social media data, large language models were adopted to quantify spatial quality characteristics including lighting, openness, cleanliness, circulation and atmosphere. The GeoDetector model was further applied to analyze the independent explanatory power and non-linear interaction mechanisms of each factor. The results showed that multimodal social media data can effectively measure spatial quality. The atmosphere factor had the strongest single-factor explanatory power. Moreover, any two-factor interaction significantly enhanced the influence of individual factors on development efficiency.