On Training Knowledge Graph Embedding Models

Knowledge graph embedding (KGE) models have become popular means for making discoveries in knowledge graphs (e.g., RDF graphs) an efficient and scalable manner. The key to success of these is their ability learn low-rank vector representations entities relations. Despite the rapid development KGE models, state-of-the-art approaches mostly focused on new ways represent embeddings interaction functions (i.e., scoring functions). In this paper, we argue that choice other training components such as loss function, hyperparameters negative sampling strategies can also substantial impact model efficiency. This area has been rather neglected by previous works so far our contribution towards closing gap a thorough analysis possible choices functions, techniques. We finally investigate effects specific scalability accuracy models.