Multi-modal entity alignment aims to identify equivalent entities between two multi-modal Knowledge graphs by integrating multi-modal data, such as images and text, to enrich the semantic representations of entities. However, existing methods may overlook the structural contextual information within each modality, making them vulnerable to interference from shallow features. To address these challenges, we propose MyGram, a \textbf{m}odalit\textbf{y}-aware \textbf{gra}ph transformer with global distribution for \textbf{m}ulti-modal entity alignment. Specifically, we develop a modality diffusion learning module to capture deep structural contextual information within modalities and enable fine-grained multi-modal fusion. In addition, we introduce a Gram Loss that acts as a regularization constraint by minimizing the volume of a 4-dimensional parallelotope formed by multi-modal features, thereby achieving global distribution consistency across modalities. We conduct experiments on five public datasets. Results show that MyGram outperforms baseline models, achieving a maximum improvement of 4.8\% in Hits@1 on FBDB15K, 9.9\% on FBYG15K, and 4.3\% on DBP15K.

Knowledge graphs (KGs) are frequently confronted with the challenge of incompleteness, a problem that extends to multimodal knowledge graphs (MKGs). The primary goal of multimodal knowledge graph completion (MKGC) is to predict missing entities within MKGs. However, current MKGC methods face difficulties in adequately addressing modal preferences and imbalances in modal information. To overcome these issues, we introduce AdaMKGC, an innovative hybrid model incorporating an adaptive modality interaction transformer. This model employs a dynamic attention interaction strategy and a self-enhancing sampling approach. AdaMKGC achieves a more precise utilization of multimodal information by integrating modal preference information into modal interactions. Additionally, it effectively mitigates the issue of modal imbalance through targeted sampling and adjustment for entities with deficient information. Experimental evaluations demonstrate AdaMKGC’s superior performance in overcoming these prevalent challenges. Compared to existing state-of-the-art MKGC models, AdaMKGC shows a notable enhancement of 28% in MR on the WN18-IMG dataset and an improvement of 2.7% in Hits@1 on the FB15k-237-IMG dataset. Our code is available at https://github.com/HubuKG/AdaMKGC .