Relation-Oriented Temporal Knowledge Graphs Completion Based on Recurrent Neural Network

Knowledge graphs (KGs) are usually incomplete, and many completion methods have emerged in the field of knowledge graphs. However, most of the current methods learn the nodes on the fixed knowledge graphs by embedding and completing directly, which is computationally complex and difficult to apply concretely. Moreover, most knowledge graphs completion methods ignore the value of temporal information. In this paper, we propose a relational-oriented temporal knowledge graphs completion method based on cyclic neural network named Rnn-Relation. Our model embeds the temporal information into a dynamic space to obtain a new relational representation vector. Then the relational information is trained by cyclic neural network to improve the relevance of relational information and temporal information. Next, the output value of the cyclic neural network layer is recoded to obtain the final value of the relational representation vector. Finally, our model utilizes the negative sample sampling to predict entities, thus significantly improving the performance of the knowledge graphs completion.