Classifying Urban Functional Zones by Integrating POIs, Place2vec, and LDA

Abstract

The urban functional zone (UFZ) is a basic unit of urban management and eco-environmental research. The study of the UFZ classification has been given more and more attention. In these studies, points of interest (POIs) are important and outstanding data, and both their semantic information and local spatial relationship information are important for the classification of UFZs. Many studies use the semantic information or local spatial relationship information of POIs for the UFZ classification, but owing to the lack of available models, only a few studies consider both of them. Combining POIs, Place2vec, and Latent Dirichlet Allocation (LDA) models, this paper proposes a new UFZ classification model that combines the semantic information and the local spatial relationship information of POIs. First, the principle and implementation of the model was described. Then the proposed model was applied to classify the UFZs of an experimental area (Chaoyang, a district of Beijing). By analyzing with manual interpretation results, accuracy of the classification results manifested that the model is useful for UFZ classification. By comparing the classification result of the proposed model with those of the Place2vec model and the LDA model, the advantage of this model was shown. Compared with the Place2vec model, this model integrates the semantic information of POIs into the representation of UFZs through the integration of the LDA model, thus improving the classification accuracy of UFZs. Compared with the LDA model, this model integrates the local spatial relationships of POIs into the representation of UFZs through the integration of the Place2vec model, thus improving the classification accuracy of UFZs.