A multi-representation knowledge graph approach for preserving and reasoning over Ming-style furniture

Ming-style furniture embodies cultural, structural and ergonomic knowledge, yet documentation remains fragmented across catalogues, standards and contemporary usage records. This study presents a multi-representation knowledge graph framework for preserving and reasoning over Ming-style furniture. An expert terminology ontology constrains large-language-model-based extraction, while fuzzy clustering and large-language-model arbitration normalize entities from museum archives, standards, e-commerce records and user feedback. The normalized knowledge is represented as aligned property-graph and semantic-web views with shared identifiers and evidence links, enabling multi-hop retrieval and logical validation. A knowledge-graph-enhanced question-answering system combines a large-language-model planner with graph query and rule-checking executors to generate evidence-grounded answers. Experiments show that the refined information-extraction module achieves an average F 1 score of 0.87, and the question-answering system reaches an expert score of 4.87/5.0, outperforming both the standard retrieval-augmented generation baseline and general graph-based baselines. The framework supports traceable, computation-driven reasoning for tangible cultural heritage.