翻译：大多前期实体对齐模型使用图卷积网络或者效果更佳的图注意力网络对实体的关系三元组进行结构建模但应用于实体对齐任务时传统静态图注意力网络的局限性主要体现在共享权重矩阵与静态性两方面。

Most early entity alignment models use graph convolutional networks or graph attention networks, which are more effective, to structurally model entity relationship triplets. However, when applied to entity alignment tasks, the limitations of traditional static graph attention networks mainly manifest in the aspects of shared weight matrices and static nature.