Improving the accuracy of knowledge graph embedding using Wikidata

Abstract

In this paper, we describe a technique for enriching the ontology of background knowledge by connecting to wikidata in order to improve the accuracy of link prediction in knowledge graphs where entities are not typed in detail. We proposed an algorithm to remove entities that are irrelevant to the type of prediction from the training as negative examples. We confirmed that the proposed algorithm improves the accuracy compared to the case without connecting to the existing wikidata.