Trajectory Similarity Learning with Auxiliary Supervision and Optimal Matching

Abstract

<jats:p>Trajectory similarity computation is a core problem in the field of trajectory data queries. However, the high time complexity of calculating the trajectory similarity has always been a bottleneck in real-world applications. Learning-based methods can map trajectories into a uniform embedding space to calculate the similarity of two trajectories with embeddings in constant time. In this paper, we propose a novel trajectory representation learning framework Traj2SimVec that performs scalable and robust trajectory similarity computation. We use a simple and fast trajectory simplification and indexing approach to obtain triplet training samples efficiently. We make the framework more robust via taking full use of the sub-trajectory similarity information as auxiliary supervision. Furthermore, the framework supports the point matching query by modeling the optimal matching relationship of trajectory points under different distance metrics. The comprehensive experiments on real-world datasets demonstrate that our model substantially outperforms all existing approaches.</jats:p>

A Graph-based Approach for Trajectory Similarity Computation in Spatial Networks

“…Yao et al [31,32] further improved the performance by devising new spatial attention mechanism and using pair-wise distance as guidance for learning. Zhang et al [35] proposed several new loss functions to improve the quality of learned embedding. All above methods are designed for similarity metrics in Euclidean space and cannot be adopted to our problem as they fail to learn the information from spatial network.…”

“…To support our task with it, we replace the grid with POIs in their framework. • Traj2SimVec [35]: This method employs a new loss for learning the trajectory similarity by point matching. We apply their model on the road network in a similar way to learn the similarity between trajectories.…”

“…As a result, the high computation cost of above similarity metrics becomes a serious problem when dealing with massive trajectory data. To resolve such problems, some recent studies [18,31,35] utilized neural network based models to learn the representation of trajectories. And the similarity between trajectories could be measured by that of the low-dimensional embedding vectors, which can be finished in linear time.…”

“…However, the difficulty in adopting deep learning based techniques to this problem is two-fold. On the one hand, it is essential to take the network structure into consideration when learning the trajectory embedding, while existing solutions for Euclidean space [18,31,35] only capture the sequence information. On the other hand, the learning process suffers from data sparsity: due to the large problem space which is exponential w.r.t.…”

Spatial-Temporal Fusion Graph Framework for Trajectory Similarity Computation

“…Trajectory similarity computation is a fundamental operation in a wide range of real world applications, such as route search [1][2][3][4][5], route planning [6][7][8], trajectory clustering [9] and transportation optimizations [10][11][12]. A Trajectory describes the path traced by bodies moving in space over time [13], and is usually represented as a sequence of discrete locations.…”