A Structure-Aware and Heterogeneous Adaptive Fusion Approach for Entity Alignment

As knowledge graphs are constantly emerging, effectively merging complex and heterogeneous knowledge graphs has become a critical challenge for improving knowledge quality and completeness. Entity alignment, as a crucial task in knowledge graph fusion, aims to identify equivalent entities across different knowledge graphs. However, existing entity alignment methods fail to fully exploit the interactions between knowledge graphs and introduce noise in graph augmentation. To address these problems, this paper proposes an entity alignment method based on structure-aware and heterogeneous adaptive fusion (SHAF). First, a hybrid graph enhancement approach based on edge reconstruction and graph propagation is introduced, where edge reconstruction optimizes the adjacency matrix, and graph propagation enhances the model's perception of latent structural features. Second, heterogeneous adaptive fusion method is designed to dynamically help the integration of entity and relation features. Specifically, gating mechanism and residual layer are employed to control the flow of information, thereby enhancing the discriminative ability of entity representations. Finally, local graph-aware contrastive optimization is developed, incorporating contrastive learning and local neighbor alignment loss to make the representations of aligned entity closer. Experimental results demonstrate that the proposed method achieves superior alignment performance on the benchmark dataset.