Geographic information extraction using natural language processing in Wikipedia texts

Geographic information extracted from texts is a valuable source of location data about documents, which can be used to improve information retrieval and document indexing. Linked Data and digital gazetteers provide a large amount of data that can support the recognition of places mentioned in text. Natural Language Processing techniques, which have evolved significantly over the last years, offer tools and resources to perform named entity recognition (NER), more specifically directed towards identifying place names and relationships between places and other entities. In this work, we demonstrate the use of NER from texts, as a way to detect relationships between places that can be used to enrich an ontological gazetteer. We use a collection of Wikipedia articles as a test dataset to demonstrate the validity of this idea. Results indicate that a significant volume of place/non-place and place-place relationships can be detected using the proposed techniques.