Extracción del contexto geográfico a apartir de NLP para información de transito en redes sociales

Abstract

Little has been said about the retrieval of spatial information from text, particularly because the term "spatial information" is associated with geometries in the form of vectors or raster-type information that express different variables or phenomena accompanied by coordinates, despite this the extraction of information in text is presented as one of the most promising advances thanks to natural language processing (NLP) and in this case it is outlined as a new field of action complementary to spatial analysis, trying to extract a specific event that happens in space and embodied in a text. The main source of text, for this research, are those shared in a collaboration network such as twitter. The extracted events are those that are found or refer to the road network and that arise in a recurring or random way, the latter, Chance, the most difficult to manage in any city that must monitor the traffic of road actors under a network of sensors that try to see the congestion of the roads and road incidents. Now, these texts were stored under a database scheme classified as road incidence that are passed over a writing pattern recognizer that extracts the location and subsequently feeds a georeferencer that returns a pair of coordinates (lat, lon), the The idea with these coordinates is to convert them into compiled data that, within a spatial analysis, show a grouping phenomenon under geostatistical techniques such as spatial autocorrelation, finding hot spots or cold spots of incident existence. The defined similar geographic results are compared with data from recent years collected by official transit entities and that are published for free access, the comparison of patterns between a previous year and those extracted with artificial intelligence show spatial behaviors and spatial self-connection preserves certain similarity revealing the usefulness of the geographical focus extraction that is proposed and possible to complement a data source for the management of road congestion and traffic incidents.