Knowledge Graph Completion Method Based on Using Multi-Hop Reasoning

In this paper, the method for knowledge graph completion based on using multi-hop reasoning is proposed. The relevance of the problem is due to the widespread use of large sparse knowledge graphs with incomplete data. Within the research, the method for restoring the structure of a knowledge graph using sequence-to-sequence multi-hop reasoning, based on the transformer architecture, is proposed. To evaluate the effectiveness of the proposed method, the study of its properties was carried out and the results were compared with the results of the existing methods used for predicting relations in the knowledge graphs. The proposed solution almost achieves the same results as methods based on embeddings. The evidential paths generated by the proposed method ensure the interpretability of the results. Our method does not rely on existing relations to create an evidential path and has the flexibility to restore missing relations in the path.