Abstract
Knowledge graph embedding approaches represent nodes and edges of graphs asmathematical vectors. Current approaches focus on embedding complete knowledgegraphs, i.e. all nodes and edges. This leads to very high computationalrequirements on large graphs such as DBpedia or Wikidata. However, for mostdownstream application scenarios, only a small subset of concepts is of actualinterest. In this paper, we present RDF2Vec Light, a lightweight embeddingapproach based on RDF2Vec which generates vectors for only a subset ofentities. To that end, RDF2Vec Light only traverses and processes a subgraph ofthe knowledge graph. Our method allows the application of embeddings of verylarge knowledge graphs in scenarios where such embeddings were not possiblebefore due to a significantly lower runtime and significantly reduced hardwarerequirements.