Tracing the spatiotemporal diffusion of cross-strait Mazu culture: a knowledge graph-driven analysis

Abstract

Mazu culture constitutes a transregional system of religious heritage characterised by ritual practices and institutional networks spanning coastal China, Taiwan, and overseas Chinese communities. Although previous studies have described its historical development and local adaptations, the relational patterns associated with its diffusion and the coexistence of symbolic continuity and local reconfiguration remain insufficiently specified. We integrate cultural-hybridity theory with knowledge-graph analysis to examine Cross-Strait spatiotemporal distributions and relational patterns. Using LLM-assisted extraction with manual verification, we constructed a Cross-Strait Mazu knowledge graph containing 871 entities and 1107 relations. The analysis identifies coastal clustering, a prominent structural position of Meizhou Ancestral Temple within the recorded spirit-division network, and relational patterns in which shared symbolic references coexist with locally differentiated practices. In a controlled 42-question comparison, LLM-only and KG+LLM achieved the same answer-level accuracy (92.9%), whereas KG+LLM showed higher claim level factual precision (94.7% vs. 85.0%) and lower rates of unsupported (3.0% vs. 10.5%) and contradictory claims (2.3% vs. 4.5%). These results provide an evidence-linked approach to examining transregional cultural heritage while distinguishing observable relational patterns from broader historical interpretation.