This is the authors' abstract. We don't add key points for this paper.
A spatio-temporal model, combining Graph Neural Networks (GNNs) and Recurrent Neural Networks (RNNs), has shown promising results in traffic flow prediction. However, existing models do not sufficiently reflect the spatio-temporal dependencies on a real road network. The observed traffic flow on a particular segment at a specific time point can be attributed to the movement of vehicles from multiple segments at various past time points; therefore, separating spatial and temporal dependencies does not precisely explain Direct Spatio-Temporal dependencies (DST-dependencies). In this paper, we introduce a Direct Spatio-Temporal graph (DST-graph) that models DST-dependencies and a novel traffic flow prediction model, named Spatio-TempoRAl dIrect GrapH aTtention network (STRAIGHT), that predicts traffic flows based on the DST-dependencies. Via extensive experiments using seven real-world datasets, we demonstrated the validity of DST-dependencies for traffic flow prediction and the effectiveness of STRAIGHT which outperformed the state-of-the-art competitors up to 37% in accuracy.
Answers are grounded in this paper and cite their sources.
No takes yet. Share an insight, caveat, or question.
Son et al. (2025) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: