A Novel Attention Method to Process Long Trajectories’ Sequences Efficiently

Processing users’ trajectories has become a crucial task in various aspects of location-based services, such as traffic prediction, trajectory recommendations, tourism recommendations, and travel planning. However, predominately most of the models that currently exist fail when they are trained with long trajectory sequences. This paper proposes a deep learning attention-based model, named (SAMO) to efficiently process the long trajectory sequences of moving objects efficiently on road networks. Indeed, (SAMO) stands for Spatial Attention Model for Objects’ Movements. The proposed model promises to process the very long trajectory sequences that will most likely be visited by the moving object whether the object’s self-history is available. In the case of no self-history, the model starts by catching the k nearest moving objects in the vicinity. In the case of self-history, the model selects trajectories like its current trip from stored history. After that, the model is trained by these objects’ trajectories either k nearest objects or similar objects and then focuses on the significant parts of these trajectories to generate results of processing and analysis of the trajectory. Overall, the proposed model outperforms competitive models by achieving up to 98% accuracy for the next multi-step prediction.