融合GAT和TransH的中韩跨语言实体对齐方法研究

Cross-lingual entity alignment in Chinese and Korean based on GAT and TransH

为研究中韩双语实体自动对齐方法,提出了一种融合图注意力网络(GAT)和基于超平面平移的知识图谱嵌入模型(TransH)的跨语言实体对齐模型.使用中韩实体数据集对模型进行验证表明,该模型的Hits@1、Hits@5和Hits@10在韩文对齐中文时分别达到了49.62%、80.89%和91.76%,在中文对齐韩文时分别达到49.79%、80.74%和91.67%,且优于传统的基于知识嵌入或图嵌入的对齐方法.因此该模型可为构建中韩对齐知识图谱以及其他语言的对齐知识图谱提供参考.

To study the automatic alignment method of Chinese and Korean bilingual entities,a cross-language entity alignment model combining graph attention network(GAT)and knowledge graph embedding model based on hyperplane translation(TransH)is proposed.Validation of the model using the Chinese and Korean entity data sets shows that the Hits@1,Hits@5,and Hits@10 of the model reached 49.62%,80.89%and 91.76%respectively,when aligning Korean to Chinese;when Chinese is aligned with Korean,it reaches 49.79%,80.74%and 91.67%respectively.It is better than traditional alignment methods based on knowledge embedding or graph embedding.Therefore,the model can provide a reference for constructing a Chinese-Korean alignment knowledge graph and alignment knowledge graph of other languages.