TAGCN: Typed Attention Graph Convolutional Networks for Entity Alignment in Cross-lingual Knowledge Graphs

Abstract

Cross-lingual entity alignment aims at integrating complementary knowledge graphs (KGs) presented in different languages. It bridges cross-lingual knowledge for knowledge discovery. In this paper, we propose a new embedding-based framework named Typed Attention Graph Convolutional Networks (TAGCN) for cross-lingual entity alignment. In TAGCN, the relation type information is fully utilized with the typed attention mechanism. Then we incorporate entity information and the relation type information of neighbors into entities through attention mechanism to iteratively learn better representation for entities. The experimental results show that our model consistently outperforms the state-of-the-art alignment methods.