FB15k-237

FB15k-237 is a widely used benchmark dataset for evaluating link prediction and knowledge completion models in knowledge graphs. The dataset is extracted from the Freebase-15k (FB15k) dataset and contains 14,541 entities, 237 unique relations, and 272,115 triples (subject-predicate-object), and is divided into training, validation, and test sets. FB15k-237 is designed to reduce the test set leakage problem caused by inverse relations, and is mainly used for link prediction and knowledge completion tasks.FB15k-237 is a widely used benchmark dataset for evaluating link prediction and knowledge completion models in knowledge graphs. The dataset is extracted from the Freebase-15k (FB15k) dataset and contains 14,541 entities, 237 unique relationships, and 272,115 triples (subject-predicate-object), and is divided into a training set, a validation set, and a test set. The FB15k-237 is designed to reduce test set leakage due to inverse relationships and is primarily used for link prediction and knowledge completion tasks.