Title: TADNM: A Transportation-mode Aware Deep Neural Model for Travel Time Estimation
Authors: Saijun Xu, Jiajie Xu, Rui Zhou, Chengfei Liu, Zhixu Li and An Liu
Abstract: Travel time estimation (TTE) has been recognized as an important problem in location-based services. Existing approaches mainly estimate travel time by learning from large-scale trajectories, they normally assume a path is in a single transportation mode (eg., driving, biking), and could not provide accurate TTE for mixed-mode paths, which are indeed common in daily life. In this paper, we propose a transportation-mode aware deep neural model called TADNM, which considers both spatio-temporal characteristics and the heterogeneity of underlying transportation modes to achieve more accurate travel time estimation. Specifically, we estimate travel time using the knowledge from (sub-)trajectories not only roughly following the target path, but also being consistent with segments of the target path in terms of transportation mode. To this end, a well-designed neural network model is proposed to integrate the rich information extracted from trajectories first, and then to learn effective representations for capturing the spatial correlations, temporal dependencies and transportation mode effects from the trajectory data. Besides, the proposed model fully considers the transition time of switching transportation mode in the path, and a transportation-mode aware attention mechanism is used to better reflect the impact of transportation mode to the required travel time. Extensive experiments on real trajectory datasets demonstrate the effectiveness of our proposed model.
