基于张量分解嵌入的时序知识图谱推理

Reasoning of temporal knowledge graphs based on tensor decomposition embedding

针对现有时序知识图谱推理中外推方法没有充分利用时间信息的问题,受张量分解模型的启发,提出将关系嵌入分为静态和动态(时序)2个部分,并通过头实体嵌入、关系嵌入和所有实体嵌入之间的双线性评分函数,计算得到对象实体的概率,从而预测对象实体。最后,在3个数据集上的实验结果验证了该方法的有效性。

In response to the insufficient utilization of temporal information in existing extrapolation methods for reasoning of temporal knowledge graphs,inspired by the tensor decomposition model,the relational embedding was divided into static and dynamic(temporal)parts.The probability of the object entity was then calculated through the bilinear scoring function between head entity embedding,relational embedding,and all entity embedding.Finally,the experimental results on three datasets verified the effectiveness of this method.