Dynamic Graph Representation Learning for Passenger Behavior Prediction

TL;DR

DyGPP addresses predicting future passenger–station interactions in subway systems by modeling passengers and stations as heterogeneous nodes within a continuous-time dynamic graph $\mathcal{G}^t$. It samples temporal interaction sequences, encodes temporal patterns with an MLP-based temporal encoder and a co-occurrence module to capture cross-node correlations, and outputs a link probability $\hat{p}_{u,s}^t$. On Beijing subway data, DyGPP achieves superior AP and AUC compared with strong baselines, validating the benefit of dynamic-graph representations for transportation analytics. The approach supports real-time, multi-step predictions and offers practical value for urban planning, crowd management, and risk mitigation by tracing evolving travel patterns. Overall, the combination of serialized history, temporal encoding, and cross-node correlation yields robust passenger behavior prediction in dynamic urban networks.

Abstract

Passenger behavior prediction aims to track passenger travel patterns through historical boarding and alighting data, enabling the analysis of urban station passenger flow and timely risk management. This is crucial for smart city development and public transportation planning. Existing research primarily relies on statistical methods and sequential models to learn from individual historical interactions, which ignores the correlations between passengers and stations. To address these issues, this paper proposes DyGPP, which leverages dynamic graphs to capture the intricate evolution of passenger behavior. First, we formalize passengers and stations as heterogeneous vertices in a dynamic graph, with connections between vertices representing interactions between passengers and stations. Then, we sample the historical interaction sequences for passengers and stations separately. We capture the temporal patterns from individual sequences and correlate the temporal behavior between the two sequences. Finally, we use an MLP-based encoder to learn the temporal patterns in the interactions and generate real-time representations of passengers and stations. Experiments on real-world datasets confirmed that DyGPP outperformed current models in the behavior prediction task, demonstrating the superiority of our model.

Figures (5)

• Figure S1: Passenger behavior prediction with a dynamic graph to trace passengers' travel patterns. This task has three dynamics: (1) travel records are infinitely growing; (2) periodic patterns over long intervals and abrupt changes over short intervals. (3) evolving relationships between passengers and stations
• Figure S2: Framework of the DyGPP model.
• Figure S3: Example of co-occurrence features.
• Figure S4: Distribution of Interaction Frequencies Between Passengers and Stations.
• Figure S5: Sequence length sensitivity in predicting.

Theorems & Definitions (2)

• Definition 1
• Definition 2