Towards Understanding the Impact of Schema on Knowledge Graph Embeddings

Knowledge graphs (KGs) enable researchers to understand a set of data within a domain of research and how different aspects of the data may connect. The methodology used to design and develop a KG varies depending on the use case. When designing a schema for a KG, also called an ontology, the developers can describe data in a rich or shallow manner. A shallow approach has uses for when there is no significant data to describe with data values, whereas a rich approach more closely mirrors reality by providing layers in the ontology to the data description. In this paper, we examine the impact that the complexity a KG schema has on their corresponding knowledge graph embeddings (KGE), where complexity varies across shallow or rich approaches for entity-to-entity relationships. We utilize the Deep Graph Library on two schemas over the same Wright State University’s CORE Scholar data. Preliminary work has shown that there are indeed differences in performance, but further investigation is needed to determine the causal mechanisms, as well as to perform additional data cleaning.