The paper introduces RotatE, a novel approach for knowledge graph embedding that models and infers various relation patterns, including symmetry/antisymmetry, inversion, and composition. RotatE represents entities as complex vectors and relations as rotations in complex vector space. Each relation is defined as a rotation from the source entity to the target entity, with the modulus of the relation embedding constrained to 1. This setup allows RotatE to effectively model all three types of relation patterns. The paper also proposes a self-adversarial negative sampling technique for efficient training. Experimental results on multiple benchmark datasets show that RotatE outperforms existing state-of-the-art models, demonstrating its effectiveness in link prediction and pattern inference.The paper introduces RotatE, a novel approach for knowledge graph embedding that models and infers various relation patterns, including symmetry/antisymmetry, inversion, and composition. RotatE represents entities as complex vectors and relations as rotations in complex vector space. Each relation is defined as a rotation from the source entity to the target entity, with the modulus of the relation embedding constrained to 1. This setup allows RotatE to effectively model all three types of relation patterns. The paper also proposes a self-adversarial negative sampling technique for efficient training. Experimental results on multiple benchmark datasets show that RotatE outperforms existing state-of-the-art models, demonstrating its effectiveness in link prediction and pattern inference.