Authors:
Kunyoung Kim
;
Donggyu Kim
and
Mye Sohn
Affiliation:
Department of Industrial Engineering, Sungkyunkwan University, Suwon, Korea
Keyword(s):
Entity Alignment, Explainable AI, Knowledge Graph, Language Model.
Abstract:
This paper proposes a Graph hierarchy and Language model-based Explainable Entity Alignment (GLEE) framework to perform Entity Alignment (EA) between two or more Knowledge Graphs (KGs) required for solving complex problems. Unlike existing EA methods that generate embedding for entities using KG structure information to calculate the similarity between entities, the GLEE framework additionally utilizes graph hierarchy and datatype properties to find entities to be aligned. In the GLEE framework, the semantically similar hyper-entities of the entities to be aligned are discovered to reflect graph hierarchy in the alignment. Also, the semantically similar datatype properties and their values of the entities are also utilized in EA. At this time, language model is utilized to calculate semantic similarity of the hyper-entities or datatype properties. As a result, the GLEE framework can trustworthy explain why the two entities are aligned using the subgraphs that consist of similar hyper
-entities, semantically identical properties, and their data values. To show the superiority of the GLEE framework, the experiment is performed using real world dataset to prove the EA performance and explainability.
(More)