Multi-modal Contrastive Representation Learning for Entity Alignment