重排序深度的最优选择研究
原标题：Optimal Re-Ranking Depth
AI 摘要
该论文研究信息检索中重排序深度（re-ranking depth）的最优选择问题，指出固定深度假设不成立，许多查询存在各自的最优重排序深度，超过该深度后效果停滞或下降。作者利用基于LLM的相关性判断进行大规模逐查询分析，发现多数标准查询性能预测（QPP）方法无法预测该深度，而基于LLM评估第一阶段排序质量的IR-DCG@10预测器可在保持整体效果的同时将平均重排序深度最多降低3倍；在最优深度选择下，MSMARCO DEV上效果可提升超7%且深度降低5倍。作者还发布了相关性判断、代码库和实验产物以支持可复现研究。
正文节选
Optimal Re-Ranking Depth Abstract. Second-stage neural rankers are commonly applied with a fixed re-ranking depth, assuming that retrieval effectiveness saturates as depth increases. Prior research has questioned the assumption that increasing the re-ranking depth yields linear performance gains, further suggesting that optimal re-ranking depth varies considerably from query to query. In the past, studying such phenomena was methodologically difficult given the scale of manual annotation require