Exploring Criteria of Loss Reweighting to Enhance LLM Unlearning