| Knowledge Graphs (KGs) are thoroughly used in artificial intelligence applications such as question-answering and recommendation systems. However, KGs are known to be incomplete. While most of the literature work focused on predicting a missing node for a given relation and an existing node, KGs can be enhanced by exploring relations between existing nodes. Such an approach is called relation prediction. We propose a relation prediction model that utilizes both KG textual and structural information. We combine walks-based embeddings with language model embeddings to represent nodes. Our model presents competitive results compared to the achievements in the relation prediction task on a widely used dataset. |
*** Title, author list and abstract as seen in the Camera-Ready version of the paper that was provided to Conference Committee. Small changes that may have occurred during processing by Springer may not appear in this window.