An Application of Natural Language Processing Techniques to Detect Spatio-Temporal Patterns of Crime as Reported in the News

The unavailability and inaccessibility of official crime data in developing countries present a major challenge for crime analysis. This study explores the potential of transforming news media content into a structured dataset suitable for pattern analysis in Georgetown, Guyana. A corpus of 5,487 newspaper articles was systematically extracted across three different newspaper platforms using crime-related keywords. In order to reveal the spatio-temporal pattern of crime in the city, the extracted news articles were analyzed using spatial point pattern methods. Findings revealed a clear temporal trend of crime peaking at 7:00 pm during weekdays and late in the night to early mornings on weekends. The temporal analysis also revealed that crime peaks on weekends. The spatial point pattern analysis confirmed significant clustering of crime at micro places, while the bivariate relationship showed the spatial dependence between pairs of crime types.