NED-GraphRAG for Evidence-Constrained Traditional Dress Design: Integrating User Needs with Cultural Knowledge Graphs

Abstract Traditional dress design based on historical culture requires evidence that can be used within a specific design context and traced back to its source. Focusing on literati gatherings in the Song dynasty, this study develops NED-GraphRAG, a GraphRAG framework that incorporates contemporary user needs into the retrieval and use of cultural evidence. User needs were identified from 3,052 public online comments using BERTopic, coding assisted by a large language model, and researcher review. The Analytic Hierarchy Process was then conducted with 18 target users to determine the relative importance of cultural elements under three design need orientations: Historical Verification, Visual Communication, and Social Ritual. Evidence drawn from historical texts, images, cultural relic materials, and modern scholarship was standardized into 2,053 evidence units and organized into a cultural evidence graph containing 4,233 nodes, 12,749 relationships, and 854 manually verified semantic triples. Six retrieval methods were evaluated using 30 test queries. NED-GraphRAG achieved an nDCG@10 of 0.821 and an AHP-AS@5 of 0.852, exceeding GraphRAG Retrieval by 0.031 and 0.104, respectively. Claim support precision reached 0.897, and the user task success rate was 92.2%. The results show that user needs can be reflected in evidence priority while the underlying cultural evidence remains unchanged. Graph relations and source controls further support the traceability of evidence used in design recommendations. The framework provides a practical method for using cultural evidence in the contemporary design of traditional dress.