Urban Mobility Prediction Using Twitter

The characteristics and dynamics of human mobility have vital implications in areas such as disaster management, transportation planning and infrastructure management. While aggregate mobility modeling is useful for getting a broader overview of the system, the prediction of future movements of people in urban areas is also of significance. This work investigates the individual-level mobility of Twitter users in three Australian cities using the concepts of entropy and predictability. Twitter users are distinguished on the basis of their movement patterns and two distinct groups are identified. The randomness and regularity in their movements are calculated via multiple metrics, and prediction for the most active users in these cities is also performed. The top 10% of Brisbane users have 76.6% prediction accuracy, much higher than the other cities, suggesting heterogeneity among various cities.