Statistical properties of urban mobility from location-based travel networks

This study explores urban mobility from a network-based perspective. The data samples used in study were collected from more than 1100 taxi drivers during a half year period in the city of Harbin in China. We extract trips from the original dataset and analyze operational efficiency. Then, by constructing travel networks based on occupied and vacant taxi trips, we calculate some statistical properties of the network such as degree, strength, edge weight, betweenness, clustering coefficient and network structure entropy. Analysis of such properties allows for a deep exploration of travel mobility. We also analyze the correlation between strength and betweenness to evaluate the importance of nodes in the network. Furthermore, two traditional community detection algorithms: the Louvain method and the visualization of similarities (VOS) method are applied to divide traffic zones in the mainland area of Harbin city. Two indices, the Rand index (RI) and the Fowkles–Mallows index (FMI) are adopted to evaluate recognition performance, which shows the similarity between administrative division and results from the algorithms. Finally, a dilatation index based on the weighted average distance among trips is applied to analyze the spatial structure of an urban city. Furthermore, hotspots are identified from local density of locations with different thresholds as determined by the Lorenz curve. •Travel trips are extracted from raw data and operation efficiency is analyzed.•Travel networks based on occupied and vacant trips are constructed.•Some statistical properties of traveling network are used to explore mobility.•Two traditional community detection algorithms are used to recognize traffic zones.•Spatial structure and hotspots are analyzed from local density of locations.