Heritage space prototypes in classical Chinese poetry

Classical Chinese poetry preserves long-term cultural memory of built space, but large-scale evidence for differences among pavilion-related built-form families remains limited. We assembled an auditable Sou-yun corpus of 371,918 poems, identified 259,342 valid references to tíng , tái , lóu , gé , and compound forms, and extracted structured semantic profiles for 247,574 poems through a rule–ML–LLM workflow. Effect-size sensitivity tests, dated Qwen reruns, DeepSeek cross-model replication, and independent blinded coding by two non-author experts supported retrieval validity and the main prototype assignments. The five prototype groups showed distinct aggregate profiles: Tíng emphasized pause and close-range garden or waterside experience; Tái emphasized elevation, ritual and commemoration; Lóu emphasized ascent, distant view and river–urban atmosphere; Gé emphasized interior refinement; and the Compound group emphasized multi-trigger scenic assemblages. A site-specific demonstration using 1195 Yellow Crane Tower records translated the Lóu profile into traceable themes for museum interpretation and digital heritage applications.