Representation learning of urban regions via mobility-signature-based zone embedding

As urbanization continues to evolve and accelerate, understanding the interactions between urban geography and large-scale mobility data has generated a great interest in the urban studies in recent years. In this paper, we present a method to learn embeddings of urban zones by utilizing the spatiotemporal characteristics of urban mobility. We extract the mobility signature from taxi trajectory data in Seoul, South Korea. Then, we apply Skip-gram model on the mobility signature to obtain the embeddings of the urban zones. Finally, we apply the spherical k-means clustering on the learned embeddings of zones to identify the urban functional regions. Through proposed approach, region maps of Seoul that can readily identify regions of similar socio-economic activities such as mobility patterns are provided.