Multi Feature Deep Fusion Based on Relationship Path Enhancement for Multi-Modal Knowledge Graph Completion

Multimodal knowledge graph integrates information from different modalities to enhance knowledge representation, but multimodal knowledge graphs still have deficiencies in completeness and accuracy. Most existing completion methods rely on entity text, images, and triplet information, ignoring the relationship path information between entities. In addition, due to the heterogeneity between text and images, there is often semantic conflict, and directly fusing information from different modalities makes it difficult to accurately capture the deep associations between text and images. Therefore, we propose a multimodal knowledge graph completion method based on relationship path enhancement with multi feature deep fusion. This method identifies and strengthens key relationship paths, while also deeply mining the deep semantic connections between multiple modalities, achieving deep integration between multiple features. Finally, the model was experimentally verified on the FB15K-237 and WN18RR datasets, results showed that the model improved on most metrics, which evidence the effectiveness of the proposed method.