通过精确切片文本提升模型对检索信息的利用效率
DeepResearch-Slice: Bridging the Retrieval-Utilization Gap via Explicit Text Slicing
- 用显式跨度预测代替隐式注意力，精准筛选有用文本
- 在6个基准上将准确率从19.1%提升至33.0%，相对提高73%
- 无需更新推理模型参数，适合快速部署到现有系统
深度研究代理主要优化搜索策略以最大化检索概率。然而，我们发现一个关键瓶颈：检索-利用差距，即模型即使检索到黄金证据也因噪声环境中的上下文盲区而无法有效使用。为此，我们提出 DeepResearch-Slice，一种简单而有效的神经符号框架。不同于隐式注意力，该方法预测精确的文本跨度索引，在推理前执行确定性硬过滤。在六个基准上的广泛评估显示显著鲁棒性提升。将该方法应用于冻结的骨干模型，准确率从19.1%提升至33.0%，相对改善73%，有效缓解噪声且无需更新推理模型参数。这些结果凸显了开放研究中显式定位机制的重要性。
原文摘要 · Abstract (English)
Deep Research agents predominantly optimize search policies to maximize retrieval probability. However, we identify a critical bottleneck: the retrieval-utilization gap, where models fail to use gold evidence even after it is retrieved, due to context blindness in noisy environments. To bridge this gap, we propose DeepResearch-Slice, a simple yet effective neuro-symbolic framework. Unlike implicit attention, our approach predicts precise span indices to perform a deterministic hard filter before reasoning. Extensive evaluations across six benchmarks show substantial robustness gains. Applying our method to frozen backbones yields a 73 percent relative improvement, from 19.1 percent to 33.0 percent, effectively mitigating noise without requiring parameter updates to the reasoning model. These results highlight the need for explicit grounding mechanisms in open-ended research.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。