Knowledge Graph Completion Based on Neighborhood-Aware Double-Layer Transformer

In knowledge graph completion, most models embed entities and relations into low-dimensional vectors and use them as inputs to learn their latent interaction features. However, these models primarily focus on static embeddings for individual triplets, neglecting the rich contextual features related to entities. This paper introduces a knowledge graph embedding model based on Neighborhood-Aware Double-Layer Transformer (NADTKE). The model consists of two layers: the bottom layer is used to learn the interaction features between the source entity and its neighborhood with respect to relations, while the top layer is responsible for aggregating contextual information from the outputs of the bottom layer. This dual-layer design effectively balances feature information from both the source entity and its neighboring entities. Experimental evaluations on the FB15k-237 and WN18RR datasets demonstrate that the proposed model achieves an MRR of 0.37 and a Hit@1 score of 0.281 on the FB15k-237 dataset, providing evidence of its effectiveness.