Marrying Query Rewriting and Knowledge Graph Embeddings.

Knowledge graph embeddings (KGEs) are useful for creating a continuous and meaningful representation of the data present in knowledge graphs (KGs). While initially employed mainly for link prediction, there has been an increased interest in querying such models using richer query languages, exploiting their continuous nature to obtain answers with a certain degree of confidence, that would not come as a result of querying KGs used to train such models. KGs can greatly benefit from having an ontology expressing conceptual knowledge of the domain and there has already been intensive research in rewriting approaches for querying the data present in KGs while taking their ontologies into account. However, these approaches have not been employed yet to query KGEs. Taking the best of both worlds, in this work we combine query rewriting in the classical DL-Lite ontology language with KGEs. More specifically, we propose a unified framework for querying KGEs that performs query rewriting for DL-Lite ontologies. We perform experiments and discuss how this combination can successfully improve query results.