Representation learning for geospatial areas using large-scale mobility data from smart card

Masanao Ochi, Yuko Nakashio, Yuta Yamashita, Ichiro Sakata, Kimitaka Asatani, Matthew Ruttley, Junichiro Mori

Abstract: With the deployment of modern infrastructures for public transit, several studies have analyzed the transition patterns of people by using smart card data and have characterized the areas. In this paper, we propose a novel embedding method to obtain a vector representation of a geospatial area using transition patterns of people from the large-scale data of their smart cards. We extend a network embedding by taking into account geographical constraints on people transitioning in the real world. We conducted an experiment using smart card data in a large network of railroads in Kansai areas in Japan. We obtained a vector representation of each railroad station using the proposed embedding method. The results show that the proposed method performs better than the existing network embedding methods in the task of multi-label classification for purposes of going to a railroad station. Our proposed method can contribute to predicting people flow by discovering underlying representations of geospatial areas from mobility data.