Semantic-Fused Multi-Granularity Cross-City Traffic Prediction

TL;DR

This work tackles data scarcity in cross-city traffic prediction by proposing Semantic-Fused Multi-Granularity Graph Transfer Learning (SFMGTL). The method combines a semantic fusion module to learn fused node embeddings and a fused graph, with hierarchical node clustering to enable multi-granularity transfer, and domain-invariant memories learned through adversarial training to mitigate negative transfer. Key contributions include the first integration of multi-semantic fusion with cross-city transfer, data-driven multi-granularity knowledge transfer, and memory-based meta-knowledge retrieval, achieving superior results on six real-world datasets with far fewer parameters than competing models. Results show that SFMGTL consistently outperforms baselines in taxi and bike demand prediction, especially during peak hours, while maintaining a compact parameter footprint and robustness to moderate source data noise. These findings highlight the practical potential of semantic fusion plus multi-granular transfer for urban traffic forecasting in data-scarce regions.

Abstract

Accurate traffic prediction is essential for effective urban management and the improvement of transportation efficiency. Recently, data-driven traffic prediction methods have been widely adopted, with better performance than traditional approaches. However, they often require large amounts of data for effective training, which becomes challenging given the prevalence of data scarcity in regions with inadequate sensing infrastructures. To address this issue, we propose a Semantic-Fused Multi-Granularity Transfer Learning (SFMGTL) model to achieve knowledge transfer across cities with fused semantics at different granularities. In detail, we design a semantic fusion module to fuse various semantics while conserving static spatial dependencies via reconstruction losses. Then, a fused graph is constructed based on node features through graph structure learning. Afterwards, we implement hierarchical node clustering to generate graphs with different granularity. To extract feasible meta-knowledge, we further introduce common and private memories and obtain domain-invariant features via adversarial training. It is worth noting that our work jointly addresses semantic fusion and multi-granularity issues in transfer learning. We conduct extensive experiments on six real-world datasets to verify the effectiveness of our SFMGTL model by comparing it with other state-of-the-art baselines. Afterwards, we also perform ablation and case studies, demonstrating that our model possesses substantially fewer parameters compared to baseline models. Moreover, we illustrate how knowledge transfer aids the model in accurately predicting demands, especially during peak hours. The codes can be found at https://github.com/zeonchen/SFMGTL.

Figures (7)

• Figure 1: Current challenges in cross-city transfer learning. (a) presents the POI similarity and DTW distance among Region 5 and other regions; (b) demonstrates dynamic spatial dependencies.
• Figure 2: The proposed model structure.
• Figure 3: Multi-granularity knowledge transfer.
• Figure 4: Training loss during pre-training and fine-tuning stages.
• Figure 5: Effects of hyperparameters on MAE.