双有界关系召回：在固定预算下超越Top-K检索的完整证据恢复
原标题：More Context, Same Budget: Dual-Bounded Relational Recall Beyond Top-K Retrieval
AI 摘要
本研究提出双有界关系召回（DBRR）方法，在固定检索预算下，通过将部分预算分配给与已选证据相关的图邻接上下文，而非仅依赖扁平top-k排序，显著提升了HotpotQA数据集上完整支持证据的恢复率。在7,405个问题上，完整证据恢复率相对匹配的扁平基线提高了23.8个百分点，其中桥接类问题提升最为显著。结果表明，在相同上下文预算下，证据的分配方式对检索完整性至关重要。
正文节选
More Context, Same Budget: Dual-Bounded Relational Recall Beyond Top-K Retrieval Abstract More context does not require a larger retrieval budget. Under the same ceiling, a retrieval system can recover more of the evidence a question requires by following relationships between evidence that flat top-k ranking leaves behind. We test that proposition with Dual-Bounded Relational Recall (DBRR), which allocates a fixed retrieval budget between relevance-selected seeds and bounded graph-adjacent cont