Resolving Top-of-Hierarchy Locations First Improves Generate-and-Rank Toponym Disambiguation

Anonymous

Abstract: Geocoding is the task of converting location mentions in text into structured geospatial data. We propose a new two-stage approach to geocoding that first resolves countries, states, and counties, and then uses these as document-level context to disambiguate the remaining location mentions. We apply this approach to two state-of-the-art geocoding models, CamCoder and SSPART. Our proposed two-stage approach to toponym resolution applied to SSPART yields state-of-the-art performance on multiple datasets. Our analysis shows that SSPART's direct incorporation of geographic database entries is key to its success over CamCoder in leveraging document context.Code and models are available at \url{https://<anonymized>}.

Paper Type: short

Research Area: Information Extraction