Abstract The morphology of traditional Chinese courtyard dwelling embodies centuries-old socio-cultural rules that are often lost in contemporary renovation and new-build projects. This study proposes an artificial intelligence assisted cultural computing framework that integrates space syntax indicators of accessibility and visual perception with convolutional and graph neural networks to reveal these tacit design rules underlying Siheyuan at scale. A dataset of 483 Siheyuan plans extracted from the 1750 Qianlong Capital Map and a measuring survey was processed through Convex Map and Visibility Graph Analysis to generate four quantitative socio-cultural indicators. A hybrid deep learning algorithm is employed to extract morphological features from the socio-cultural indicators, which are represented in the form of images and graphs. The extracted feature vectors are reduced to two dimensions, and the samples are clustered into 9 groups based on an elbow analysis. Based on correlation and comparative analyses on the samples of the 9 groups, latent design rules underlying Siheyuan morphology are identified. The resulting two-dimensional feature map offers conservation architects a quantitative, similarity-based reference system for renovation and culturally legible new design of Siheyuan. The framework is transferable to other building types, which advances architectural morphology from ideal type description to data-driven pattern recognition.
Wang et al. (Fri,) studied this question.