Discovery of urban functional regions based on Node2vec

Identifying urban functional regions is a trending topic in urban computing. It helps understand the economic and cultural development of cities and assists decision-makers in land-use planning. However, the studies to date have not fully mined the spatio-temporal characteristics of location data, and most have used the direct clustering method to judge differences among urban land use types, resulting in low identification accuracies. Here, we propose a novel framework for urban functional region discovery based on residents’ travel patterns. The framework uses GPS trajectory data and check-in data to construct a travel pattern graph with temporal and spatial attributes and learns the vector representations of urban regions using the Node2vec method to identify urban functional regions. In addition, a novel detection method that combines semantic information of check-in data and well-known points of interest (POIs) was constructed to annotate the semantics of the functional regions. Real location data sets were used for verification. The experimental results showed that the proposed graph embedding learning method could effectively discover urban functional regions. Moreover, this method solved the problem of accurately identifying the semantics of functional regions to a certain extent.