Construction of Chinese Traditional Culture Knowledge Graph and Intelligent Question Answering System Based on Graphormer

Traditional graph neural networks have limitations in modeling long-range multi-hop relation paths and sparse semantic entity representations when constructing knowledge graphs for Chinese traditional culture. To address variant character forms, syntactic abbreviations, and sparse entity semantics in ancient texts, this paper proposes a knowledge graph construction and question-answering system based on Graphormer. The Confucian four-element inheritance structure of “teacher-student-writing-school-event” and the three-element mapping of “rites, textiles, and crafts” are formalized as node types and directed relations in a knowledge graph. An edge-type encoding and path-constraint mechanism is constructed accordingly. Based on this structure, Graphormer’s shortest-path distance bias and edge-feature encoding are guided to prioritize multi-hop paths consistent with Confucian transmission logic. By embedding the philosophical transmission structure into a graph attention mechanism, the method achieves collaborative modeling of sparse semantic relations and long-range dependencies. Experiments show that the proposed method achieves Hits@10 of 0.659 in the knowledge completion task and accuracies of 0.773, 0.715, and 0.667 in four-hop multi-hop question answering, verifying the effectiveness of the structural expression and reasoning paradigm.