Classifying Urban Functional Zones by Integrating Place2Vec and GCN

Abstract

Urban functional zones play a crucial role in urban management and planning. The study on urban functional zone classification has become a current hotspot. In these studies, point of interest (POI) is an important data source that contains valuable social, economic, cultural, and geographic information. Classifying urban functional zones based on the spatial association features of POIs is an important direction of current studies. However, when obtaining the vectors of POI types, these studies considered the local spatial association of POIs, but when obtaining the vectors of urban functional zones, they ignored the spatial structure of POIs. To solve this problem, this study proposed a new urban functional zone classification method by integrating the Place2Vec and the graph convolutional network (GCN) models. In the proposed method, the Place2Vec model was used to obtain the vectors of POI types based on the local spatial association of POIs; the GCN model was used to integrate the spatial structure of POIs with the vectors of POI types to calculate the vectors of urban functional zones for urban functional zone classification. The proposed method was used to classify the urban functional zones in simulation datasets and in Chaoyang District of Beijing. The classification accuracies of the proposed method were compared with those of the Place2Vec model. The results showed that the proposed method had higher classification accuracies than the Place2Vec model.