RotatE is a knowledge graph embedding method that models relations as rotations in complex vector space. The method represents entities as complex vectors and relations as rotations between entities. Each relation is defined as a rotation from the source entity to the target entity in the complex vector space. The model can infer and model various relation patterns, including symmetry/antisymmetry, inversion, and composition. A novel self-adversarial negative sampling technique is proposed for efficient training of the model. Experimental results on multiple benchmark knowledge graphs show that RotatE outperforms existing state-of-the-art models for link prediction. The model is scalable and can effectively model and infer all three types of relation patterns. RotatE also performs well on a benchmark explicitly designed for composition pattern inference and modeling. The model's ability to implicitly represent relation patterns is demonstrated through analysis of relation embeddings. The model is evaluated on several datasets, including FB15k, WN18, FB15k-237, WN18RR, and Countries. Results show that RotatE achieves state-of-the-art performance on all benchmarks. The model is also compared with other knowledge graph embedding methods, including TransE, DistMult, ComplEx, HolE, and ConvE. The results indicate that RotatE outperforms these models in terms of link prediction accuracy. The model's ability to handle complex relation patterns is further validated through experiments on the Countries dataset, where it successfully infers composition patterns. The model's performance is also evaluated on different relation categories, showing that it performs well on non-injective relations, especially many-to-many relations. The model is also compared with a probabilistic framework for knowledge graph embedding, showing that it outperforms the corresponding model in terms of performance. The model's ability to handle uncertainties in knowledge graphs is also discussed, with plans to further explore this area in future work.RotatE is a knowledge graph embedding method that models relations as rotations in complex vector space. The method represents entities as complex vectors and relations as rotations between entities. Each relation is defined as a rotation from the source entity to the target entity in the complex vector space. The model can infer and model various relation patterns, including symmetry/antisymmetry, inversion, and composition. A novel self-adversarial negative sampling technique is proposed for efficient training of the model. Experimental results on multiple benchmark knowledge graphs show that RotatE outperforms existing state-of-the-art models for link prediction. The model is scalable and can effectively model and infer all three types of relation patterns. RotatE also performs well on a benchmark explicitly designed for composition pattern inference and modeling. The model's ability to implicitly represent relation patterns is demonstrated through analysis of relation embeddings. The model is evaluated on several datasets, including FB15k, WN18, FB15k-237, WN18RR, and Countries. Results show that RotatE achieves state-of-the-art performance on all benchmarks. The model is also compared with other knowledge graph embedding methods, including TransE, DistMult, ComplEx, HolE, and ConvE. The results indicate that RotatE outperforms these models in terms of link prediction accuracy. The model's ability to handle complex relation patterns is further validated through experiments on the Countries dataset, where it successfully infers composition patterns. The model's performance is also evaluated on different relation categories, showing that it performs well on non-injective relations, especially many-to-many relations. The model is also compared with a probabilistic framework for knowledge graph embedding, showing that it outperforms the corresponding model in terms of performance. The model's ability to handle uncertainties in knowledge graphs is also discussed, with plans to further explore this area in future work.