Context-based Natural Language Processing for GIS-based Vague Region Visualization

Vernacular regions such as central Ohio are popularly used in everyday language; but their vague and indeterministic boundaries affect the clarity of communicating them over the geographic space. This paper introduced a context-based natural language processing approach to retrieve geographic entities. Geographic entities extracted from news articles were used as location-based behavioral samples to map out the vague region of central Ohio. Particularly, part of speech tagging and parse tree generation were employed to filter out candidate entities from English sentences. Propositional logic of context (PLC) was introduced and adapted to build the contextual model for deciding the membership of named entities. Results were automatically generated and visualized in GIS using both symbol and density mapping. Final maps were consistent with our intuition and common sense knowledge of the vague region.