MobiCLR: Mobility Time Series Contrastive Learning for Urban Region Representations

TL;DR

MobiCLR addresses the challenge of deriving robust urban region representations from dynamic mobility data by introducing a mobility time-series contrastive learning framework. It leverages three encoders to capture inbound, outbound, and combined mobility semantics, an instance-level contrastive objective, and an auxiliary regularizer to align flow-specific and holistic region patterns, trained on two weeks of taxi data. The approach yields superior predictions of educational attainment, income, and the Social Vulnerability Index across Chicago, New York, and Washington D.C. and demonstrates cross-city transferability. This work advances urban computing by integrating temporal mobility dynamics with multi-indicator prediction, enabling more informed city planning and resilience analyses.

Abstract

Recently, learning effective representations of urban regions has gained significant attention as a key approach to understanding urban dynamics and advancing smarter cities. Existing approaches have demonstrated the potential of leveraging mobility data to generate latent representations, providing valuable insights into the intrinsic characteristics of urban areas. However, incorporating the temporal dynamics and detailed semantics inherent in human mobility patterns remains underexplored. To address this gap, we propose a novel urban region representation learning model, Mobility Time Series Contrastive Learning for Urban Region Representations (MobiCLR), designed to capture semantically meaningful embeddings from inflow and outflow mobility patterns. MobiCLR uses contrastive learning to enhance the discriminative power of its representations, applying an instance-wise contrastive loss to capture distinct flow-specific characteristics. Additionally, we develop a regularizer to align output features with these flow-specific representations, enabling a more comprehensive understanding of mobility dynamics. To validate our model, we conduct extensive experiments in Chicago, New York, and Washington, D.C. to predict income, educational attainment, and social vulnerability. The results demonstrate that our model outperforms state-of-the-art models.

Figures (5)

• Figure 1: Overall framework of MobiCLR. We first pre-process region-wise hourly inbound and outbound trips. Subsequently, a two-step data augmentation process is applied to the mobility time series data, followed by the feeding of this augmented data into three encoders: $f_\theta^i$, $f_\theta^o$, and $f_\theta^io$. Here, $f_\theta^i$ and $f_\theta^o$ learn inbound- and outbound-specific human mobility patterns in each region, respectively, while $f_\theta^{io}$ extracts semantics containing both inbound- and outbound-specific features. After pre-training, the embedding vectors obtained through $f_\theta^{io}$ are utilized in downstream applications.
• Figure 2: Graphical illustration of data augmentation strategy.
• Figure 3: Results of social vulnerability prediction under different time series data augmentation strategies, including individual transformations and sequential compositions of two transformations. The diagonal entries represent the results of single transformations, while the off-diagonal entries show the results of compositions of two transformations.
• Figure 4: Sensitivity analysis
• Figure 5: $R^2$ for the transferability test on social vulnerability index. The model is trained on one source city and then evaluated on other target city.