Integration of Textual and Structural Information for Knowledge Graph Completion

Knowledge graph completion (KGC) aims to infer missing entities based on observed entities, which is of great significance for many downstream applications, such as information retrieval. Graph Neural Networks (GNNs) have been widely applied in KGC due to their powerful graph modeling capabilities. They learn information from neighboring nodes by passing messages between adjacent nodes. However, the textual information of entities and explicit relational information in the knowledge graph are rarely utilized in existing KGC methods, which typically focus only on the structural information of the graph in GNN-based approaches. T Therefore, a model integrating textual and structural information is proposed for knowledge graph completion (ITS-KGC). ITS-KGC integrates pre-trained language models to model entity descriptions while utilizing GNN to aggregate structural information from different relations. Finally, it explicitly updates relation embeddings through a gated recurrent unit (GRU), thereby enhancing the model's capacity to understand diverse relationships. Extensive experiments were conducted on the FB15K-237 and WN18RR datasets to validate the performance of the proposed model. Compared to baseline models, ITS-KGC achieved outstanding performance in the evaluation metrics of MRR and Hits@n.