| Abstract |
During the process of modern industrialization and urbanization, it has become one of the major daily activities that people drive private cars to fulfill their travel demands. The trajectory data generated during the usage of private cars plays as the intuitive embodiment of people's travel behavior. In particular, the stay behavior, i.e., people need to stay and take time carrying out their own activities when they drive to a specific location, contains crucial information for understanding users' travel behavior and mobility motivations. In this paper, via leveraging the private car trajectory data, we strive to propose a novel approach to percept and predict stay of interests, called SOI. The goal is to predict the stay interest of a private car user will stay to a given location, this is important information for vehicle services such as travel semantic analysis and smart recommendation service. Specifically, we first propose a stay behavior perception method to detect stay behavior from large-scale private car trajectory dataset. Then, we design a spatiotemporal factor extraction method considering the spatial aggregation, time period and spatiotemporal similarity correlation of stay behaviors, which can reduce the sparsity and non-stationary problems of stay behavior data. Furthermore, we propose a prediction method based gradient boosting decision trees to estimate the future stay interest of private car users' stay behavior. We conduct extensive experiments based on the real-life private car trajectory dataset. For the stay interest prediction of stay behavior, achieve prediction precision of 0.89 and recall of 0.85. To the best of authors' knowledge, this is the first work in literature that exploits private car trajectory data and discovers the stay behavior of private car users and hence provides new perspective for understanding people's travel behavior. © 2019 IEEE. |
