<p>With the advent of the big data era, the rapid growth and diversity of information has led to the growing problem of incompleteness in knowledge graphs(KGs). Therefore, Knowledge Graph Completion (KGC) has garnered a focal point of recent research efforts. However, nowadays, most of the KGC uses the low-dimensional embedding of relations and entities to capture the interaction features, but most of the KGC techniques are difficult to capture the real interaction features between different entities and relations, and many hidden features are not captured. The present paper proposes a KGC based on cross-attention of multiple concepts with a view to capturing more hidden features. Specifically, multiple semantic feature representations are obtained by projecting the low-dimensional vectors of entities and relations. Then, each semantic feature is innovatively divided into multiple implicit features, which are more likely to capture hidden complex features. The cross-attention module employs sophisticated feature extraction to capture the interactions between features of entities and relations, as well as the internal hidden features of entities and relations, without compromising the complexity of the model. Extensive experimental results on multiple standard link prediction benchmark datasets demonstrate that the method proposed in this paper demonstrates competitive performance in comparison to other comparative methods. Our public implementation is available at github.com/xiaodong20182018/MRCCA2025.</p>