Adversarial Attack against Cross-lingual Knowledge Graph Alignment

• DOI: 10.18653/v1/2021.emnlp-main.432
• Publication Date: 2021-12-17T03:56:42Z
• Authors: Zeru Zhang; Zijie Zhang; Yang Zhou; Lingfei Wu; Sixing Wu; Xiaoying Han; Dejing Dou; Tianshi Che; Da Yan
• Keywords: Vulnerability; Maximization

Recent literatures have shown that knowledge graph (KG) learning models are highly vulnerable to adversarial attacks. However, there is still a paucity of vulnerability analyses cross-lingual entity alignment under This paper proposes an attack model with two novel techniques perturb the KG structure and degrade quality deep alignment. First, density maximization method employed hide attacked entities in dense regions KGs, such derived perturbations unnoticeable. Second, signal amplification developed reduce gradient vanishing issues process attacks for further improving effectiveness.