Geo-FuB: A Method for Constructing an Operator-Function Knowledge Base for Geospatial Code Generation Tasks Using Large Language Models

Geo-FuB, a framework for constructing a geospatial operator-function knowledge base using large language models, addresses coding issues in geospatial code generation. It integrates retrieval-augmented generation with an external knowledge base, comprising Geo-FuSE for semantic feature extraction, Geo-FuST for operator combination identification, and Geo-FuM for aligning these with the function semantic framework. Evaluated on 154,075 Google Earth Engine scripts, Geo-FuB achieves 88.89% overall accuracy, with 92.03% structural and 86.79% semantic accuracy. This framework enhances the reliability of large language models in geospatial code generation, offering a solution to challenges like "coding hallucinations" and improving the integration of domain-specific expertise.