COOL: A Conjoint Perspective on Spatio-Temporal Graph Neural Network for Traffic Forecasting

TL;DR

Con Conjoint Spatio-Temporal graph neural network is proposed, which models heterogeneous graphs from prior and posterior information to conjointly capture high-order spatio-temporal relationships and proposes a conjoint self-attention decoder that models diverse temporal patterns from both multi-rank and multi-scale views.

Abstract

This paper investigates traffic forecasting, which attempts to forecast the future state of traffic based on historical situations. This problem has received ever-increasing attention in various scenarios and facilitated the development of numerous downstream applications such as urban planning and transportation management. However, the efficacy of existing methods remains sub-optimal due to their tendency to model temporal and spatial relationships independently, thereby inadequately accounting for complex high-order interactions of both worlds. Moreover, the diversity of transitional patterns in traffic forecasting makes them challenging to capture for existing approaches, warranting a deeper exploration of their diversity. Toward this end, this paper proposes Conjoint Spatio-Temporal graph neural network (abbreviated as COOL), which models heterogeneous graphs from prior and posterior information to conjointly capture high-order spatio-temporal relationships. On the one hand, heterogeneous graphs connecting sequential observation are constructed to extract composite spatio-temporal relationships via prior message passing. On the other hand, we model dynamic relationships using constructed affinity and penalty graphs, which guide posterior message passing to incorporate complementary semantic information into node representations. Moreover, to capture diverse transitional properties to enhance traffic forecasting, we propose a conjoint self-attention decoder that models diverse temporal patterns from both multi-rank and multi-scale views. Experimental results on four popular benchmark datasets demonstrate that our proposed COOL provides state-of-the-art performance compared with the competitive baselines.

Figures (6)

• Figure 1: Transitional patterns in traffic networks could be diverse, varying in terms of variances (a) and periodicity (b).
• Figure 2: Framework overview of the proposed COOL.
• Figure 3: Hyperparameter study of the proposed COOL on PEMS08 and METR-LA.
• Figure 4: Visualization of prediction results on METR-LA.
• Figure 5: Visualization of learned attentions scores of multi-scale attention module. Sub-figure (a) is the attention scores of window size 4, Sub-figure (b) is the attention scores of window size 3, Sub-figure (c) is the attention scores of window size 2.