Entity Alignment (EA) aims to identify equivalent entities across different knowledge graphs (KGs), a crucial task for knowledge fusion. Recent approaches treat EA as a representation learning challenge through graph embedding. However, they do not fully exploit the most reliable structural information to encode the KGs and leave considerable room for further exploration of decoding strategies. To address these issues, we propose Context-Aware and Multi-View Enhanced Model for Entity Alignment (CMEA), which concentrates solely on graph structure for both encoding and decoding in entity alignment. Specifically, we take a data augmentation strategy by adding reverse edges and constructing twin graphs. In the encoding phase, we design a Twin Graph Convolutional Network to model entities by leveraging the semantic context of neighboring entities and their associated relations. In the decoding phase, we utilize multi-view adjacency matrices to derive fine-grained entity representations without training, thereby improving EA performance. Comprehensive experiments conducted on four real-world datasets highlight the advantages of our method over previous approaches.
原文链接
分享
引用
