Entity–Relation Joint Extraction from Urban Planning Standards without Annotated Data Resources

Urban planning standards serve as critical guidelines for global urban development, yet the inefficiency of their knowledge services severely hinders the intelligent processes of planning formulation and evaluation. Existing generic knowledge extraction techniques struggle to adapt to highly specialized urban planning standard texts due to terminology‐intensive content, complex inter‐entity relations, and scarce annotated resources. To address these challenges, this study proposes a knowledge extraction method that combines large language models (LLMs) with conventional small models. Twenty‐three entity types and 35 relation types were defined, and an urban planning standards ontology was constructed. With annotation rules designed, an LLM‐based annotated corpus, including 4905 sentences and 17,056 triples, was constructed. The Urban Plan‐Bert‐Global Pointer‐based Linking (UPB‐GPLinker) was built for entity–relation joint extraction based on a customized dictionary and a pretrained Urban Plan‐Bert (UPB) model. Compared with the baseline model, UPB‐GPLinker significantly improves the F1 score by at least 2.48% and the Recall by at least 0.37%, verifying the feasibility and effectiveness of the improved method. This study fills the gap of knowledge extraction in urban planning standards and lays a foundation for forming standard knowledge graphs and providing intelligent services.