Many recent studies of Entity Alignment (EA) use Graph Neural Networks (GNNs) to aggregate the neighborhood features of entities and achieve better performance. However, aligned entities in real Knowledge Graphs (KGs) usually have non-isomorphic neighborhood structures due to the different data sources of KGs. Therefore, it is insufficient to simply compare the global direct neighborhood of aligned entities, which may also become a variable for the EA judgment. In this paper, we propose a Relation-based Adaptive Neighborhood Matching method ( RANM ), which matches larger range and higher confidence neighborhoods for aligned entities based on relation matching instead of alignment seeds. RANM first uses alignment seeds to construct the best relation matching set, and then performs local direct neighborhood matching and feature aggregation on the candidate alignments. To obtain high-quality entity embeddings, we design a variant attention mechanism based on heterogeneous graphs, which considers the heterogeneity of relations in KGs. We also adopt a bi-directional iterative co-training to further improve the performance. Extensive experiments on three well-known datasets show our method significantly outperforms 14 state-of-the-art methods, and is 3.01-11.5% higher than the best-performing baselines in Hits@1. RANM also shows high performance on the long-tailed entities and the dataset with less alignment seeds.
The success of current entity alignment (EA) tasks largely depends on the supervision information provided by labeled data. Considering the cost of labeled data, most supervised methods are challenging to apply in practical scenarios. Therefore, an increasing number of works based on contrastive learning, active learning, or other deep learning techniques have been developed, to solve the performance bottleneck caused by the lack of labeled data. However, existing unsupervised EA methods still face certain limitations; either their modeling complexity is high or they fail to balance the effectiveness and practicality of alignment. To overcome these issues, we propose a learnable convolutional attention network for unsupervised entity alignment, named LCA-UEA. Specifically, LCA-UEA performs convolution operations before the attention mechanism, ensuring the acquisition of structural information and avoiding the superposition of redundant information. Then, to efficiently filter out invalid neighborhood information of aligned entities, LCA-UEA designs a relation structure reconstruction method based on potential matching relations, thereby enhancing the usability and scalability of the EA method. Notably, a similarity function based on consistency is proposed to better measure the similarity of candidate entity pairs. Finally, we conducted extensive experiments on three datasets of different sizes and types (cross-lingual and monolingual) to verify the superiority of LCA-UEA. Experimental results demonstrate that LCA-UEA significantly improved alignment accuracy, outperforming 25 supervised or unsupervised methods, and improving by 6.4% in Hits@1 over the best baseline in the best case.
Entity Alignment (EA) in Knowledge Graphs (KGs) is a crucial task for the integration of multiple KGs, facilitating the amalgamation of multi-source knowledge and enhancing support for downstream applications. In recent years, unsupervised EA methods have demonstrated remarkable efficacy in leveraging graph structures or utilizing auxiliary information. However, the increasing complexity of many modeling methods limits their applicability to large KGs in real-world scenarios. Given that most EA encoders primarily focus on modeling one-hop neighborhoods within the KG’s graph structure while neglecting similarities among multi-hop neighborhoods, we propose an efficient and effective unsupervised EA method, MPGT-Align, based on a multi-hop pruning graph transformer. The core innovation of MPGT-Align lies in mining multi-hop neighborhood features of entities through two components: Pruning-hop2Token and Attention-based Transformer encoder. The former aggregates only those multi-hop neighborhoods that contribute to alignment targets, inspired by search pruning algorithms. The latter empowers MPGT-Align to adaptively extract more effective alignment information from both entity itself and its multi-hop neighbors. Furthermore, Pruning-hop2Token serves as a non-parametric method that not only reduces model parameters, but also allows MPGT-Align to be trained with small batch sizes, thereby enabling efficient handling of large KGs. Extensive experiments conducted across various benchmark datasets demonstrate that our method consistently outperforms most existing supervised and unsupervised EA techniques.