Using Generic Ontologies to Infer the Geographic Focus of Text

Christos T. Rodosthenous, Loizos Michael

Abstract: Certain documents are naturally associated with a country as their geographic focus. Some past work has sought to develop systems that identify this focus, under the assumption that the target country is explicitly mentioned in the document. When this assumption is not met, the task becomes one of inferring the focus based on the available context provided by the document. Although some existing work has considered this variant of the task, that work typically relies on the use of specialized geographic resources. In this work we seek to demonstrate that this inference task can be tackled by using generic ontologies, like ConceptNet and YAGO, that have been developed independently of the particular task. We describe GeoMantis, our developed system for inferring the geographic focus of a document, and we undertake a comparative evaluation against two freely-available open-source systems. Our results show that GeoMantis performs better than these two systems when the comparison is made on news stories whose target country is either not explicitly mentioned, or has been artificially obscured, in the story text.