A ritual-aware cognitive-adaptive knowledge graph for the Dongba pictographic script heritage

Dongba script, the world’s only living pictographic writing system, lacks datasets organized at the ritual event level and addressing expert–novice cognitive asymmetry. We present RACK-Dongba, a four-layer neuro-symbolic framework integrating: (1) AHP-based user needs analysis revealing complete non-overlap between expert (n = 7) and novice (n = 18) top-3 priorities; (2) NS-DPOnto ontology extending CIDOC CRM with a RitualEvent semantic layer; (3) CLIP-ViT-B/32 cross-document matching and GLM-4V multimodal annotation; and (4) a SWRL-based cognitive load classifier with dual-layer Neo4j query templates. The DPKG dataset comprises 1788 character entities, 3 ritual event nodes (481 ritual-affiliated), and cognitive load labels (HIGH: 161, MEDIUM: 1081, LOW: 546). Experiment 1 achieves F1 = 77.7%, outperforming CIDOC-KG (44.0%) and BERT-KG (15.8%). Experiment 2 (n = 32, NASA-TLX) validates classification efficacy (F(2,62) = 363.298, p < 0.001, partial η² = 0.921). RACK-Dongba is the first semantic knowledge graph for the Dongba script and the first pictographic heritage system to operationalize cognitive needs as a design principle with dual-layer audience-adaptive access.