Despite the widespread use of Knowledge Graph Embeddings (KGE), little is\nknown about the security vulnerabilities that might disrupt their intended\nbehaviour. We study data poisoning attacks against KGE models for link\nprediction. These attacks craft adversarial additions or deletions at training\ntime to cause model failure at test time. To select adversarial deletions, we\npropose to use the model-agnostic instance attribution methods from\nInterpretable Machine Learning, which identify the training instances that are\nmost influential to a neural model's predictions on test instances. We use\nthese influential triples as adversarial deletions. We further propose a\nheuristic method to replace one of the two entities in each influential triple\nto generate adversarial additions. Our experiments show that the proposed\nstrategies outperform the state-of-art data poisoning attacks on KGE models and\nimprove the MRR degradation due to the attacks by up to 62% over the baselines.\n