Using Twitter for Next-Place Prediction, with an Application to Crime Prediction

This research focuses on two problems. First, we investigate the prediction of social media users' spatial trajectories in the real world. Recent work on this task focused on utilizing cellular network traces and location-based social network services such as Foursquare, all of which emit structured geospatial information (e.g., cellular tower identifiers, GPS coordinates, and venue identifiers). Less attention has been paid to the rich textual content that users often publish in tandem with the structured information. From information in social media, we use two ways to define individual's movement pattern: the nearest venue type and the minimum distance to each venue type. We conclude that the combination of textual content together with a user's current location could be used to predict his or her movement pattern. We investigate methods of integrating textual content into existing next-place prediction models, and we demonstrate a significant improvement in next-place prediction compared to several baselines derived from published research. Second, we examine the correlation between routine activity extracted from next-place predictions and the occurrence of crimes in a major United States city, with the goal of aiding future research into automatic crime prediction. The result of next-place prediction with application in crime prediction will help policy making in police's law enforcement. The results support our hypothesis that people's movement patterns are correlated with crime rates. Then we have also built a classification solution to test how the movement trajectories help crime prediction. We extract the count of occupants who move to each venue type as features to demonstrate the movement trajectories. We build a classification solution to predict whether there is any crimes or not in a past time period. This preliminary classification model just utilize the density of people's movement to predict the happening of crimes, which need further investigation with more features like historical crime density.