Unlocking the Power of Large Language Models for Entity Alignment