翻译：实体对齐任务中知识图谱中的实体节点与其邻接节点通常具有较大差异而对于不同类型的节点应用共享的权重矩阵将导致模型难以正确区分实体节点与邻接节点从而降低模型的表示能力。因此本文分别使用两个不同的权重矩阵对不同实体节点进行计算修改后的注意力参数公式如下式所示。

In entity alignment tasks, the entity nodes in knowledge graphs and their adjacent nodes often have significant differences. Applying a shared weight matrix to different types of nodes will make it difficult for the model to correctly distinguish entity nodes from adjacent nodes, thereby reducing the model's representation ability. Therefore, this paper calculates different entity nodes using two different weight matrices, and the modified attention parameter formula is shown in the following equation.