Mobility Knowledge Graph and its Application in Public Transport

Sammanfattning: Efficient public transport planning, operations, and control rely on a deep understanding of human mobility in urban areas. The availability of extensive and diverse mobility data sources, such as smart card data and GPS data, provides opportunities to quantitatively study individual behavior and collective mobility patterns. However, analyzing and organizing these vast amounts of data is a challenging task. The Knowledge Graph (KG) is a graph-based method for knowledge representation and organization that has been successfully applied in various applications, yet the applications of KG in urban mobility are still limited. To further utilize the mobility data and explore human mobility patterns, the included papers constructed the Mobility Knowledge Graph (MKG), a general learning framework, and demonstrated its potential applications in public transport.Paper I introduces the concept of MKG and proposes a learning framework to construct MKG from smart card data in public transport networks. The framework captures the spatiotemporal travel pattern correlations between stations using both rule-based linear decomposition and neural network-based nonlinear decomposition methods. The paper validates the MKG construction framework and explores the value of MKG in predicting individual trip destinations using only tap-in records.Paper II proposes an application of user-station attention estimation to understand human mobility in urban areas, which facilitates downstream applications such as individual mobility prediction and location recommendation. To estimate the 'real' user-station attention from station visit counts data, the paper proposes a matrix decomposition method that captures both user similarity and station-station relations using the mobility knowledge graph (MKG). A neural network-based nonlinear decomposition approach was used to extract  MKG relations capturing the latent spatiotemporal travel dependencies. The proposed framework is validated using synthetic and real-world data, demonstrating its significant value in contributing to user-station attention inference.