Impact of entity names embeddings on the quality of entity alignment

Abstract

The problem of merging multilingual knowledge graphs (KG) is becoming more and more relevant. The main step for its solution is the identification of equivalent entities and their descriptions. It is also known as the entity alignment (EA) problem. In recent years, EA methods based on embeddings of entities have been actively studied. Recent studies show that the quality of these approaches depends on how information about the structure of knowledge graphs and methods for constructing embeddings of entity names are used. This article presents experiments, the purpose of which is to improve the alignment of entities on the English-Russian dataset.