Deriving the geographic footprint of cognitive regions

Abstract

The characterization of place and its representation in current Geographic Information System (GIS) has become a prominent research topic. This paper concentrates on places that are cognitive regions, and presents a computational framework to derive the geographic footprint of these regions. The main idea is to use Natural Language Processing (NLP) tools to identify unique geographic features from User Generated Content (UGC) sources consisting of textual descriptions of places. These features are used to detect on a map an initial area that the descriptions refer to. A semantic representation of this area is extracted from a GIS and passed over to a Machine Learning (ML) algorithm that locates other areas according to semantic similarity. As a case study, we employ the proposed framework to derive the geographic footprint of the historic center of Vienna and validate the results by comparing the derived region against a historical map of the city.