Improving Land Use Classification using Human Mobility-based Hierarchical Place Embeddings

Understanding land use patterns is becoming increasingly important for effective urban planning. Meanwhile, place embeddings, which are generated from human mobility data collected from mobile devices, have become a popular method to understand the functionality of places, and are essential for various downstream tasks including land use classification and mobility prediction. Place embeddings with high spatial resolution are desirable for land use classification, however, downscaling the spatial resolution could deteriorate the quality of embeddings due to data sparsity, especially in less populated areas. We address this issue by proposing a method that is able to generate fine grained place embeddings, by leveraging spatial hierarchical information according to the local density of observed data points. We demonstrate the practical value of the generated fine grained place embeddings to better understand land use, using real world trajectory data from 3 cities in Japan and comparing the proposed method with the baseline non-hierarchical method. Our technique of incorporating spatial hierarchical information can complement and reinforce various place embedding generation methods.