Multi-modal based region representation learning considering mobility data in Seoul

Abstract

There are limits to simply understanding the complex form of modern cities. Recently, research for expressing various characteristics of urban areas using methods such as a graph embedding technique is being conducted. Most research utilizes a variety of data that exist in cities, but for mobility data, typically only one type of data is used, such as a taxi. In this study, we intend to perform multi-modal based region representation learning that can reflect various mobility data. Multi-modal is the simultaneous use of multiple results of a single-modal learned from each mobility data to find characteristics of urban areas through different aspects of mobility data. In addition, this study not only considers various types of mobility data but also tries to identify various characteristics of urban areas by classifying transportation user types. Based on the results applied to the actual Seoul area in the experiments, it is found that the results using the multi-modal outperform other models and the single-modal. In addition, the importance of classifying transportation user types is presented, and the impact of each user type is analyzed and presented.