Objective The integrated knowledge system of “environment – architecture – behaviour – perception” embedded in classical Chinese gardens has not been systematically recognized in the contemporary research system dominated by the study of material entities and combined with the contemporary design theory system, resulting in the disconnection between ontology research and design research of classical gardens, and the inefficiency of the transformation and application of research results. Aiming at the problem of excavating and characterizing multi-dimensional elements in classical gardens, this research reviews imperial poems in the Fuhai area of Yuanmingyuan Garden, summarizes the coupling relationship of key elements, and puts forward a suitable method of knowledge characterization, so as to expand the research path of “immaterial” gardens depicted in the poems.

Methods The utilization of text mining, knowledge graph and associated technologies, in conjunction with the Forty Scenes of Yuanmingyuan and the Yangshi Lei Archives, along with other historical materials, facilitates the characterization of the imperial poetry texts regarding the Fuhai area in Yuanmingyuan during the Qianlong period. This characterization employs literary cartography and semantic network analysis to elucidate the linkage mechanism between the objective material elements in the ten scenes of the Fuhai area and the subjective behavioral perceptions. Initially, the aforesaid poems are subject to the steps of physical lexicon production, word division, word annotation, and data cleaning. The four types of elements, namely architecture, environment, behavior and perception, are then extracted. These elements are then combined with the 3D spatial model to create a layout map and to map the local semantic network corresponding to the physical object and the spatial layout. Secondly, the four types of elements are automatically recognized by software, supplemented by manual recognition and correction, to obtain the strength of semantic association between the elements, establish the binary matrix relationship, and form the Node-Edge reflecting the interrelationships of the words and phrases. The semantic information mapped by the words and their relative positions in the Node-Edge is then used to carry out clustering, and the connectivity of the elements is determined according to the semantic and syntactic associations to form the associated element library. The final stage of the analysis involves the clustering of behavioral and perceptual information according to different identity consciousnesses and emotional tendencies. Three typical identities are selected for identity space mapping: Emperor, literati and loiterer. This is done so that the “environment – architecture” elements associated with “perception – behavior” under each typical identity can be excavated. The material space elements associated with “perception – behavior” under each typical identity are then explored. The spatial patterns under the three types of identities are constructed from the levels of spatial element, organizational structure, and behavioral perception, combined with the spatial information in the Yangshi Lei Archives and the Forty Scenes of Yuanmingyuan.

Results The research has constructed a multi-dimensional elements graph of "environment - architecture - behavior - perception" in the Fuhai area, and has restored the spatial usage scenarios of Fuhai through spatial semantic analysis. The “behavior – perception” characteristics of three typical identity consciousnesses — emperor's responsibility, literati’s mentality, and loiterer's posture — are extracted, and the corresponding spatial semantic network of the garden is constructed.

Conclusion The research proposes a novel methodology for the analysis of classical gardens, with a particular focus on the identification of spatial information and subjective perception data embedded within textual, visual, and physical elements. The proposed methodology involves a systematic exploration of the relationship between behavior, perception, environment, and architecture, with the objective of enhancing the understanding of the multifacetedness of these cultural spaces. The methodology outlined in this research addresses the challenges associated with the classification, correlation, and representation of garden information, offering a comprehensive framework for the analysis of classical gardens. The research sheds light on the spatial design principles employed during the reign of the Qianlong Emperor, providing a detailed insight into the construction of classical gardens. The research on the single-level coupling elements of traditional gardens is broken, and the synergistic response mechanism of “environment – architecture – behaviour – perception” is constructed, thereby overcoming the bottleneck of information synthesis and multi-objective computation regarding classical gardens. This provides methodological support for the cognition of classical gardens and quantification of multi-source information.