关键词: 实体对齐, 知识嵌入, 注意力机制, 依存句法分析, Bert

Abstract: Entity Alignment is to identify the same entity in the Knowledge Graph of different sources. Most of the existing Entity Alignment models focus on the characteristics of the entity itself. Some models introduce entity relationships and attribute information to assist in alignment. However, these methods ignore the potential neighborhood infor-mation and semantic information in the entity. In order to solve the above problems, a multi-view entity alignment model combining triples and text attributes is proposed. This model divides entity information into multiple views to achieve alignment. For the lack of neighborhood information, Graph Convolutional Neural network and translation model are used to learn the relationship information of embedded entities in parallel, aiming at the lack of semantic information. Word embedding and pre-trained language model are used to learn the semantic information of attribute text. The experimental results show that the proposed method is always superior to other Entity Alignment methods on multiple benchmark datasets, and is expected to improve the alignment accuracy and discrimination ability.

Key words: entity alignment, knowledge embedding, attention mechanism, dependency syntactic parsing, Bert