Deconstruction and Reconstruction of Intangible Cultural Heritage Symbols Driven by GraphSAGE: Taking the Intelligent Generation of Bohai Mohe Embroidery Topological Patterns as an Example

Guanhong Zhan

Article
2026 / Volume 9 / Pages 549-570
Received 10 June 2025; Accepted 8 August 2025; Published 6 March 2026
https://doi.org/10.31881/TLR.2026.549

Abstract
Traditional methods struggle to capture the complex structure and cultural semantics of Bohai Mohe embroidery, leading to distortion in intelligent pattern generation. This study proposes a graph sample and aggregate (GraphSAGE)-based framework that models embroidery patterns as graphs, encoding visual symbols as nodes and connections. A two-layer GraphSAGE network extracts local stitching and global cultural features, followed by feature fusion and dynamic edge adjustment to generate structurally sound, culturally consistent patterns. Exper-iments on a custom dataset show high pattern symmetry consistency (0.881) and semantic fidelity (up to 0.970), demonstrating the method’s effectiveness in digitalizing intangible cultural heritage through structured modeling and multilevel feature perception.

Keywords
GraphSAGE framework, embroidery pattern, visual symbol, cultural feature, semantic fidelity