Social media as a predictor of urban crime

Abstract

This study attempts to determine the importance of analyzing crimes at increasingly precise spatial and temporal levels. Messages from the social media source Twitter are used to predict the distribution of crimes in Montreal by estimating the actual population in the area and characterizing it by mood. Multi-level Poisson models are used to predict violent and property crimes for a particular street segment according to the time of day. The results show that any crime analysis in Montreal must take into account variance in crime at the microsite level and incorporate intraday periods. Also, the characterization of a city's actual population has been identified as a promising avenue for crime prediction. This study suggests that the analysis of Twitter data makes it possible to draw some conclusions, but it still needs to be further developed.