The field of graph neural networks (GNNs) is witnessing significant advancements in both their application and the understanding of their vulnerabilities. Recent research has focused on enhancing the robustness and efficiency of GNNs in various applications, including fraud detection, social behavior analysis, and financial risk analysis. A notable trend is the development of sophisticated attack methodologies that exploit the inherent vulnerabilities of GNNs, such as inference attacks, backdoor attacks, and adversarial attacks. These studies not only highlight the potential risks associated with deploying GNNs in sensitive areas but also propose innovative solutions to mitigate these risks. For instance, novel frameworks have been introduced to improve the imperceptibility and effectiveness of attacks, leveraging advanced techniques like prompt-based inference, subgraph triggers, and homogeneous node injection. On the application side, there is a growing emphasis on semi-supervised learning approaches to overcome the challenges posed by the scarcity of labeled data, particularly in fraud detection. These approaches utilize temporal transaction graphs and risk propagation models to enhance the detection of fraudulent activities with minimal labeled data. Overall, the field is moving towards a deeper understanding of GNNs' capabilities and limitations, with a strong focus on developing more secure and efficient models for real-world applications.

Noteworthy Papers

• Prompt-based Unifying Inference Attack on Graph Neural Networks: Introduces ProIA, a framework that enhances inference attack capabilities on GNNs by retaining crucial topological information during pre-training.
• Attack by Yourself: Effective and Unnoticeable Multi-Category Graph Backdoor Attacks with Subgraph Triggers Pool: Proposes EUMC, a method for conducting multi-category graph backdoor attacks using category-aware subgraph triggers.
• Semi-supervised Credit Card Fraud Detection via Attribute-Driven Graph Representation: Presents GTAN, a semi-supervised GNN that outperforms state-of-the-art baselines in fraud detection with minimal labeled data.
• Hypergraph Attacks via Injecting Homogeneous Nodes into Elite Hyperedges: Introduces IE-Attack, a novel framework for enhancing the performance and imperceptibility of hypergraph attacks.
• Unveiling the Threat of Fraud Gangs to Graph Neural Networks: Multi-Target Graph Injection Attacks against GNN-Based Fraud Detectors: Develops MonTi, a transformer-based model for multi-target graph injection attacks, showcasing superior performance over existing methods.