Relation Transformation based on Graph Convolution Network for Entity Alignment

Abstract

Entity alignment has lately received great attention for its key role in knowledge fusion, which has a goal that links entities with the same meaning from different knowledge graph. Most recently, many novel methods based on graph convolution network (GCN) have emerged as interesting models for entity alignment. Although these existing methods can further improve the quality of entity embedding, they ignored the influence of the embeddings of multiple relationships between entities on entity alignment. In addition, current models do not considerably learn the embeddings of relationship. To address these problems, we adopt a novel Relation Transformation based on Graph Convolution Network for entity alignment, named RT-GCN. Specifically, we point out a new GCN to aggregate the information of entities, which aims to strikingly obtain entities embeddings. Moreover, we develop a novel relational transformation method to generate relational embeddings. The crucial role of relational transformation is to strengthen the process of alignment and improve the robustness of the model. Experimental results on three datasets state that the proposed model RT-GCN performs surprisingly better than the state-of-the-art methods.