Seek：面向知识检索的自评估探索框架
原标题：Seek: Self-Evaluative Exploration for Knowledge Retrieval
AI 摘要
该论文提出 Seek，一个无需训练的迭代式知识检索框架，通过在测试时多轮交互语料库来克服单次检索的局限。每轮由 LLM 生成伪段落扩展查询、检索器获取新候选、专用评估器给出分级相关性判断并指导后续轮次。在 TREC Deep Learning 上 Seek 匹配训练过的重排序器并提升 Recall@100；在 BRIGHT 上 Qwen2.5-7B 相对 BM25 提升 82%，GPT-4.1 达到 37.4 平均 nDCG@10，超过最强基线 37%。代码、提示和数据已公开。
正文节选
Seek: Self-Evaluative Exploration for Knowledge Retrieval Abstract. LLM-based retrievers and rerankers have advanced passage ranking, yet both paradigms interact with the corpus in a single pass and commit to the resulting candidate set, leaving relevant documents permanently unrecoverable once missed. We introduce Seek, Self-Evaluative Exploration for Knowledge Retrieval), a training-free framework that addresses this limitation through iterative corpus interaction at test time. At each round,