Tag
Proposes M2GDT, a novel MKGC framework that uses an MLLM-guided diffusion transformer with relation-adaptive mixture-of-experts to align and denoise multimodal features, outperforming baselines on three benchmark datasets.