Benchmark datasets for knowledge graph completion experiments: FB15k, WN18, FB15k-237, and WN18RR.

In the knowledge graph completion experiment, multiple benchmark datasets were adopted, which were divided into training sets, validation sets and test sets. The specific descriptions are as follows:(1) FB15k: FB15k is a subset of the World Knowledge Base Freebase. It is a representative knowledge graph completion task dataset containing various real-world knowledge facts, which can evaluate the performance of models when dealing with diverse entity-relationship data.(2) WN18: WN18 is a subset of the language knowledge base WordNet. It is a knowledge graph dataset representing lexical relationships between different words, such as superior-subordinate relationships and subordination relationships, and is used to evaluate the performance of models in handling semantic relationships.(3) FB15K-237: FB15K-237 is a subset of the FB15k dataset, with all reversible relationships removed. It is a more challenging dataset for knowledge graph completion tasks.(4) WN18RR: WN18RR is a subset further filtered from the WN18 dataset, with all reversible relationships removed, making the dataset more challenging. It is mainly used for knowledge graph completion tasks.