KGCBC: Knowledge graph completion model of BERT hybrid Convolution

As a critical foundational resource supporting numerous intelligent tasks, knowledge graphs face the challenge of incomplete triples. Common Knowledge Graph Completion (KGC) methods are typically divided into structural embeddingbased models and semantic comprehension-based models, which often risk partial information loss. This paper proposes KGCBC, a hybrid model integrating BERT and convolutional networks, designed to extract structural and semantic information from triples through dedicated convolutional and BERT modules. The convolutional module incorporates multi-head attention mechanisms to dynamically enhance structural features. Ablation studies validate the complementary roles of the two modules. Evaluations on benchmark datasets demonstrate KGCBC’s outstanding performance, surpassing state-of-the-art models on WN18RR.