Trajectory Similarity Assessment On Road Networks Via Embedding Learning

Abstract

Trajectory similarity assessment is a basic task in trajectory data analysis and application, such as trajectory clustering, route planning and POI recommendation. However, the evaluation of similar trajectories in urban environments often suffer from road network constraints, sampling rate difference and GPS errors, etc. To solve these problems, we propose a trajectory similarity assessment approach based on road network embedding, which can capture both topology and spatiality of road networks for embedding learning. Specifically, it performs random walk to spatial query through depth-first search and introduces distance to optimize the loss function. When mapped to road networks, trajectory embedding can be obtained and the similarity of trajectories can be evaluated. Experiments on real data show that our approach is robust and efficient for trajectory similarity assessment.