Trajectory Similarity Learning with Fusion of Spatial and Structural Feature Difference

The application of mobile devices and sensing equipment is driving the generation of a substantial volume of trajectory data. Calculating trajectory similarity is one of the most fundamental challenges in simplifying the analysis of trajectory data. To effectively integrate spatial and structural features of trajectories, we propose a novel robust model based on feature differential fusion for trajectory similarity calculation. Specifically, the difference function allows the model to distinguish the differing components of spatial and structural features, enabling the effective fusion of trajectory features. Meanwhile, owing to the inherent use of attention mechanisms within the Transformer, the model is capable of capturing deep-level correlations within trajectories. Furthermore, as each trajectory is embedded as a d-dimensional vector, it significantly accelerates the computation of trajectory similarity compared to approximate algorithms. Finally, we conducted training and testing on two widely used real trajectory datasets, Geolife and Porto, and the results thoroughly demonstrate the performance and effectiveness of our model.