Effect of enriched ontology structures on RDF embedding-based entity linking

Abstract

RDF embeddings are recently used in Entity Linking systems for disambiguation of candidate entities to match the best mention and entity pairs. In this study, we evaluate the effect of enriched ontology structures for disambiguation task when RDF embeddings are used to identify semantic relatedness between knowledge base concepts. We generate a domain-specific core ontology and put new components upon previous ontology structures. In this way, we obtain four different enriched structures and transform them into RDF embeddings. Then, we observe which enriched structure has more importance to enhance the overall performance of RDF embeddings-based Entity Linking approaches. We select two well-known knowledge-base-agnostic approaches, including AGDISTIS and DoSeR and adapt them into RDF embeddings-based entity disambiguation. Finally, a domain-specific evaluation dataset is generated from Wikipedia to observe the effect of enriched structures on these adapted approaches.