Position-Aware Structural Knowledge Sharing-Based Federated Graph Learning for Intelligent Transportation Systems

Federated Graph Learning (FGL) combines the powerful graph data modeling capabilities of Graph Neural Networks (GNNs) with the distributed processing requirements of intelligent transportation systems (ITS), making it a prominent focus in recent research. In the context of the Internet of Everything (IoE), diverse devices and services generate highly heterogeneous, non-independent, and non-identically distributed (non-IID) data, which limits model generalization and training efficiency. To tackle these challenges, this paper proposes a Position-Aware Structural Sharing Federated Graph Learning Framework tailored for ITS applications. This framework enhances GNNs’ capacity to process cross-domain graph data, significantly improving model applicability and performance across various ITS scenarios. Specifically, we use a structural encoder alongside a position-aware structural encoder to capture generic structural knowledge, sharing these embeddings across clients in an FGL setup. Extensive experiments on different datasets demonstrate that our method outperforms existing approaches in handling non-IID federated graph learning tasks, particularly within cross-dataset and cross-domain environments.