Classifying Urban Functional Zones by Integrating the Homogeneity and Structural Similarity of POIs

Abstract

The classification of urban functional zones helps to understand the spatial structure of urban functions. However, achieving high classification accuracy is challenging, because of the complexity and randomness of urban components and their spatial distribution. An important idea in existing urban functional zone classification research is to extract the spatial context of points of interest (POIs) based on their spatial relationships, embed the POI representations, and then calculate the vector representation of the urban functional zones for classification. Important features that could be extracted from the spatial relationships between the POIs are the homogeneity and structural similarity between POIs. Although their important impact on urban functional zone classification has received attention, most studies construct urban functional zone classification methods that are based on the homogeneity of POIs. Limited studies combine them to construct urban functional zone classification methods. In this paper, a novel method for urban functional zone classification was proposed by integrating the homogeneity and structural similarity between POIs. First, the neighborhood structure of the POIs is defined based on the spatial relationships between POIs. Second, biased random walking strategies are designed to generate the spatial contexts of POIs. Third, the spatial contexts of the POIs are input into the Skip-Gram model to generate vectors for the POI types. Then, the vectors of urban functional zones are calculated. Finally, the support vector machine (SVM) is used for urban functional zone classification that is based on their vectors. The model was applied in Chaoyang District, Beijing. The feasibility, accuracy, and stability of the model were verified by comparing it with the Place2vec and the DeepWalk models.