LLM-RankFusion: Mitigating Intrinsic Inconsistency in LLM-based Ranking