更多上下文，相同预算：超越Top-K检索的双边界关系召回
More Context, Same Budget: Dual-Bounded Relational Recall Beyond Top-K Retrieval
- Radiant Institute for Manifold Studies (RIMS)（辐射流形研究所（RIMS））
机构由 AI 辅助整理，请以论文原文为准。
AI总结：
该研究提出双边界关系召回（DBRR）方法，在相同检索预算下通过分配相关性种子与图相邻上下文，使HotpotQA支持证据恢复率提升23.8个百分点，优于纯Top-K检索。
AI中文摘要：
更多上下文并不需要更大的检索预算。在相同的上限下，检索系统通过利用纯Top-K排名所遗漏的证据间关系，可恢复问题所需的更多证据。我们通过双边界关系召回（DBRR）验证该假设，该方法在相关性选择的种子与有界图相邻上下文之间分配固定检索预算，同时使用相同的相关性排名阶段、相同的最大检索单元数和token数，与匹配的纯Top-K检索进行对比。结果是每个问题都能完整恢复HotpotQA官方的支持证据集。在7405个FullWiki问题中，Primary DBRR分配相比匹配的纯基线方法，将完整支持证据恢复率提高了23.8个百分点（配对风险差为0.2377；问题级自举95%区间为0.2269至0.2489）。它在1952个问题上表现提升，5261个问题上表现相当，192个问题上表现受损。桥接问题驱动了该效果，提升幅度为28.7个百分点；对比问题则显示出较小的4.2个百分点差异。在预先指定的仅用于评估的诊断总体中，真实关系也优于随机邻居和保持度的打乱图对照组。结果明确：在相同上下文预算下，完整证据检索不仅取决于排名最高的项，还取决于上下文在这些项周围的分配方式。关系分配恢复了纯Top-K检索所遗漏的完整证据集。
英文摘要：
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 context, against matched flat top-k retrieval using the same relevance-ranking stage and the same maximum number of retrieval units and tokens. The outcome is complete recovery of the official HotpotQA supporting-evidence set for each question. Across 7,405 FullWiki questions, the Primary DBRR allocation increased complete supporting-evidence recovery by 23.8 percentage points over its matched flat baseline (paired risk difference 0.2377; question-level bootstrap 95% interval 0.2269 to 0.2489). It improved 1,952 questions, tied on 5,261, and harmed 192. Bridge questions drove the effect, with a 28.7-point increase; comparison questions showed a smaller 4.2-point difference. In a prespecified, evaluation-only diagnostic population, real relationships also outperformed random-neighbor and degree-preserving shuffled-graph controls. The result is straightforward: under the same context budget, complete-evidence retrieval depends not only on which items rank highest, but on how context is allocated around them. Relational allocation recovered complete evidence sets that flat top-k retrieval left incomplete.
