Cross-lingual entity alignment in multilingual environments faces numerous challenges,particularly when long-tail entity neighbors are sparse and structural information is limited,which restricts alignment accuracy.To address this issue,a method integrating both structural and semantic features for cross-lingual entity alignment is designed.Specifically,semantic em-beddings are generated using a multilingual pre-trained model as the initial entity features.Subsequently,structural features are extracted through a graph neural network,and a high-speed gating mechanism is employed to integrate both semantic and struc-tural information,enhancing the discriminative ability of entity representations.Experimental results demonstrate that on cross-lingual datasets ZH-EN,JA-EN,and FR-EN,the Hits@1 evaluation metrics reached 0.858,0.918,and 0.941,respectively,signifi-cantly outperforming traditional baseline models and showcasing superior alignment performance.
万方数据