Knowledge Graph Completion Fusing Description and Structural Feature

Knowledge Graph Completion (KGC) aims to complement the structure of knowledge graphs by predicting missing entities or relations within them. Currently, mainstream KGC methods only leverage either the structural information of triples or the semantic information of knowledge graphs, lacking effective approaches to fuse these two types of information. Currently, mainstream KGC methods only leverage either the structural information of triples or the semantic information of knowledge graphs, lacking effective approaches to fuse these two types of information. Model (SSFM) for knowledge graph completion that integrates semantic and structural information. SSFM employs the pre-trained language model BERT to SSFM employs the pre-trained language model BERT to encode the descriptive information of entities and relations to obtain the semantic information of the knowledge graph. It uses GAT to mine the structural information of the knowledge graph and designs an attention-based feature fusion module to effectively integrate semantic and structural information for model training to achieve KGC. Detailed experiments were conducted on the public datasets FB15k-237 and WN18RR. Results show that the model exhibits superior performance in terms of the Hits@3 metric compared to baseline models. On the FB15k-237, the Hits@3 metric of this model is 1.3% higher than that of the SACN baseline model; on the WN18RR, this metric is 3.6% higher than that of the SACN baseline model.