Graph Neural Networks for Vehicular Social Networks: Trends, Challenges, and Opportunities

TL;DR

This survey addresses how Graph Neural Networks (GNNs) can advance Vehicular Social Networks (VSNs) within Intelligent Transportation Systems. It surveys GNN architectures (GCNN, RGNN, GAT, GGNN) and learning tasks (node/graph classification, link prediction, community detection, graph generation) in the context of VSNs, and catalogues applications across traffic flow, trajectory forecasting, TF, TSC, driving assistance, VRP, TDA, and connectivity. The work highlights that hybrid GNNs and attention-based methods often outperform traditional baselines, while noting a critical gap: no study yet models a complete, standalone VSN with all functional components. It also documents open datasets, trends, and future directions—emphasizing Real-time scalability, multimodal data integration, privacy/security, and collaboration-driven datasets to enable practical, large-scale VSN implementations. Overall, GNNs are poised to play a central role in real-time, socially aware, large-scale VSNs, provided the field overcomes dataset, deployment, and privacy challenges.

Abstract

Graph Neural Networks (GNNs) have emerged as powerful tools for modeling complex, interconnected data, making them particularly well suited for a wide range of Intelligent Transportation System (ITS) applications. This survey presents the first comprehensive review dedicated specifically to the use of GNNs within Vehicular Social Networks (VSNs). By leveraging both Euclidean and non-Euclidean transportation-related data, including traffic patterns, road users, and weather conditions, GNNs offer promising solutions for analyzing and enhancing VSN applications. The survey systematically categorizes and analyzes existing studies according to major VSN-related tasks, including traffic flow and trajectory prediction, traffic forecasting, signal control, driving assistance, routing problem, and connectivity management. It further provides quantitative insights and synthesizes key takeaways derived from the literature review. Additionally, the survey examines the available datasets and outlines open research directions needed to advance GNN-based VSN applications. The findings indicate that, although GNNs demonstrate strong potential for improving the accuracy, robustness, and real-time performances of on task-specific or sub-VSN graphs, there remains a notable absence of studies that model a complete, standalone VSN encompassing all functional components. With the increasing availability of data and continued progress in graph learning, GNNs are expected to play a central role in enabling future large-scale and fully integrated VSN applications.

Figures (14)

• Figure 1: The structure of this survey paper.
• Figure 2: Vanilla architecture of a GNN showing the basic steps of transforming the input graph into embeddings.
• Figure 3: Illustration of the different types of embeddings: Node Embedding, Edge Embedding, and Graph Embedding. Each embedding results in a low-dimensional space representation.
• Figure 4: Visualization of the embedding process where the original 3-D network is transformed into an embedding space with 2-D vectors by the means of an encoding function $ENC()$.
• Figure 5: Illustration of the graph neural network embedding two nodes $u$ and $v$ using their attribute vector $x_{u,0}$ and $x_{v,0}$.