The paper discusses the increasing recognition of Heterogeneous Graph Neural Networks (HGNNs) in areas like web and e-commerce and the necessity for resilience against adversarial attacks. The authors present HGAttack, the first dedicated gray box evasion attack method for heterogeneous graphs. The design of HGAttack includes a novel surrogate model and the use of gradient-based methods for perturbation generation. The surrogate model effectively uses heterogeneous information by extracting meta-path induced subgraphs and applying GNNs to learn node embeddings with distinct semantics from each subgraph. This approach improves the transferability of generated attacks on the target HGNN and significantly reduces memory costs. The paper validates the efficacy of HGAttack through comprehensive experiments on three datasets.
Publication date: 19 Jan 2024
Project Page: Unknown
Paper: https://arxiv.org/pdf/2401.09945
