End-to-end Relation-Enhanced Learnable Graph Self-attention Network for Knowledge Graphs Embedding

The knowledge graphs embedding performance of the classic graph convolutional network has been limited due to the large-scale knowledge information. The complex knowledge information requires the model for better learnability rather than linearly weighted qualitative constraints. By studying the structural characteristics of the knowledge graph and investigation the imbalance of knowledge information, the end-to-end relation-enhanced learnable graph self-attention network for knowledge graphs embedding is proposed in this work. A relation-enhanced adjacency matrix is constructed to take into account the incompleteness of the knowledge graph. The convolutional knowledge sub-graph is introduced by using a graph self-attention network to obtain the entity node information’s global encoding and relevance ranking. The training effect of the Convolutional Knowledge Base (convKB) model is improved by changing the construction of negative samples, and a better reliability score in the decoder can be obtained. The experimental results on the data set FB15k-237 and WN18RR show that the proposed method achieves better scores than the compared methods in terms of Hits@10 and MRR.