Knowledge graph embedding (KGE) has become crucial for knowledge-based applications like question-answering systems. In multimodal knowledge graphs, integrating heterogeneous modalities to capture semantic relationships is particularly challenging. While multimodal knowledge graph embedding (MKGE) models have been developed, few focus on leveraging relational semantics across modalities. To bridge this gap, we propose BiMFEA, a representation learning model that integrates multimodal data using a fusion gate and entity structure attention mechanism. By refining entity and relation representations with BiComplex embeddings and balancing contributions from multimodal fusion and structural attention, BiMFEA ensures robust entity modeling. Experiments on benchmark datasets show BiMFEA’s state-of-the-art performance for MKGE tasks.