Detecting geospatial location descriptions in natural language text

References to geographic locations are common in text data sources including social media and web pages. They take different forms from simple place names relative expressions that describe location through a spatial relationship reference object (e.g. the house beside Waikato River). Often complex, multi-word phrases employed road railway cross at right angles; line with canal) where relationships communicated various parts of speech prepositions, verbs, adverbs adjectives. We address problem automatically detecting geospatial descriptions, which we define as those include relation terms referencing objects, distinguishing them non-geographical descriptions book on table). experiment several methods for automated classification expressions, using features machine learning bag words detect distinctive words, word embeddings encode meanings manually identified language patterns characterise expressions. Using three sets created this study, find ensemble meta-classifier approaches, variously combine predictions other classifiers features, provide best F-measure 0.90