The article proposes a method for developing universal knowledge graph embeddings by fusing large knowledge graphs based on the owl:sameAs relation. This ensures each entity has a unique identity, enhancing the consistency of entity representation across different sources. The approach is implemented using DBpedia and Wikidata, generating embeddings for a large number of entities, relations, and triples. An API is also developed to provide these embeddings as a service. The experiment results indicate that these universal embeddings encode better semantics compared to single knowledge graph embeddings.
Publication date: 23 Oct 2023
Project Page: https://github.com/dice-group/UniversalEmbeddings
Paper: https://arxiv.org/pdf/2310.14899
