基于实体活跃度及复制生成的时序知识图谱推理

现有时序知识图谱推理主要是基于静态知识图谱的推理方法，通过知识图谱的结构特征挖掘潜在的语义信息和关系特征，忽略了实体时序信息的重要性，因此提出一种基于实体活跃度及复制生成机制的时序知识图谱推理方法（EACG）。首先，通过改进的图卷积神经网络对多关系实体建模，有效挖掘知识图谱的潜在语义信息和结构特征。其次，时序编码器基于实体活跃度学习实体的时序特征。最后，使用复制生成机制进一步学习知识图谱的历史信息，提升对时序数据建模的能力。在时序知识图谱数据集ICEWS14、ICEWS05-15、GDELT上推理的实验结果表明，EACG在MRR评估指标中分别优于次优方法2%、10%和5%。

Temporal knowledge graph prediction based on entities activity and copy generation

The existing temporal knowledge graph reasoning methods were mainly based on static knowledge graph reasoning methods. These methods utilized the structural features of knowledge graphs to mine potential semantic information and relationship features, ignoring the importance of entity temporal information. Therefore, this paper proposed an based entity activity and copy generation(EACG) temporal knowledge graph reasoning method. Firstly, this paper employed an improved graph convolutional network to model multi-relational entities, effectively mining the latent semantic information and structural features of knowledge graphs. Next, this paper utilized the temporal encoder to learn the temporal characteristics of entities based on the activity of entities. Finally, this paper used the copy generation mechanism to further learn the historical information of knowledge graphs and improve the ability to model temporal data. The experimental results of reasoning on temporal knowledge graph datasets ICEWS14, ICEWS05-15, and GDELT show that EACG outperforms the sub-optimal method by 2%, 10% and 5% respectively in the MRR evaluation index.