Extracting hidden information from human mobility patterns is one of the long-standing challenges of urban studies. In addition, exploring the relationship between urban functional structure and traffic spatial interaction pattern has long been of interest. Recently, vehicle GPS trajectory data emerged as a popular data source for revealing human mobility patterns and urban functions. However, few studies have fully leveraged traffic interaction information that is hidden in human mobility patterns to identify urban functions at the road segment level. To address this issue, a geo-semantic analysis framework was introduced in this study to model the relationship between traffic interaction and urban functions at the road segment level. First, a Road-Trajectory corpus was built and trained to obtain the semantic embedding representation of road segments. Then, considering topological connections between road segments, we used a graph convolutional neural network model to process the contextual and topological information to classify social functions along streets. A case study in Beijing, China, using a large volume of real-world taxi trajectories data, was conducted. The results show that our proposed methods, with relative less loss and high accuracy, outperform other comparative methods for classifying urban functions at the road segment level. This work contributes to the assessment of urban functional structure, and further aiding urban planners in designing better urbanization strategies with regard to traffic interaction and urban space structure.

Graph convolutional neural network, Semantic embedding, Traffic interaction patterns, Urban functions, Vehicle GPS trajectory data

South China Normal University