Construction and Application of Knowledge Graph for Bridge Inspection

Abstract

This paper constructs a knowledge graph and complex question answering corpus for bridge inspection, which lays the foundation for constructing an automatic question answering system oriented to the characteristics of this field. Many related works are currently presented for the open domain due to the characteristics of massive scale, high data quality, and easy availability of general domain knowledge graph data, but there is only a little research on domain knowledge graph construction and complex question answering. This paper proposes a naming method of “component name_(center pile number)” to distinguish the same components of different bridges. Then given the unique tree structure of the knowledge graph for bridge inspection, we creatively propose a three-level relational knowledge graph construction method and annotate datasets containing five types of complex question answering. Finally, we use our corpus to conduct experiments on mainstream complex question answering models and further demonstrate the effectiveness of the domain-specific knowledge graph construction method.