Generating a Semantic Parsing Dataset for GeoKBQA Over OpenStreetMap

This study tries to automatically construct a GeoKBQA dataset to transition to more comprehensive, neural-based GeoKBQA. The current GeoKBQA datasets are lacking in coverage, efficiency, scalability, and compatibility as there are only two datasets available that are manually constructed. This pipeline yields a substantial and diverse dataset comprising 1,776 entries with various logical forms, featuring a wide range of geo-functions and schema item compositions, including functions, classes, and relations.