Knowledge Graph Augmentation with Entity Identification for Improving Knowledge Graph Completion Performance

A knowledge graph often lacks some existent triples. Knowledge graph completion is a technique for complementing such triples and its performance can be improved by augmenting triples from other external databases. However, entity names often differ between the original knowledge graph and an external database, which reduce the augmentation’s efficiency. In this study, we identify the same entities that have different names (orthographic variants) that come from different sources, merge them into one entity, and augment the knowledge graphs. Our proposed method exploits in the original knowledge graph and the external database the similarity of triples, which were embedded using BERT. Experimental evaluation on our knowledge graph completion performance showed that our proposed method with graph information effectively outperformed two baselines.