User-station attention inference using smart card data: a knowledge graph assisted matrix decomposition model

Abstract Understanding human mobility in urban areas is important for transportation, from planning to operations and online control. This paper proposes the concept of user-station attention, which describes user’s (or user group’s) interest or dependency on specific stations. The contributes a better understanding (e.g., travel purposes) facilitates downstream applications, such as individual prediction location recommendation. However, intrinsic unsupervised learning characteristics untrustworthy observation data make it challenging estimate real attention. We introduce attention inference problem using station visit counts public transport develop matrix decomposition method capturing simultaneously similarity station-station relationships knowledge graphs. Specifically, captures information matrix. It extracts stations’ latent representation hidden relations (activities) between stations construct graph (MKG) smart card data. neural network (NN)-based nonlinear approach extract MKG spatiotemporal dependencies. case study uses both synthetic real-world validate proposed by comparing with benchmark models. results illustrate significant value contributing inference. model improves estimation accuracy 35% MAE 16% RMSE. Also, not sensitive sparse provided only positive observations are used.