# DeepResearch-Slice: Bridging the Retrieval-Utilization Gap via Explicit Text Slicing

Source: [DeepResearch-Slice: Bridging the Retrieval-Utilization Gap via Explicit Text Slicing](<https://www.sota2.com/research/paper/695f39f053fdd5c361743ec5>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-01-09T05:00:05.326Z

Authors: Shuo Lu, Yinuo Xu, Jianjie Cheng, Lingxiao He, Meng Wang, Jian Liang

Published: 2025-12-16T07:07:28.000Z

## Abstract

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.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Multi-Hop Question Answering on HotpotQA (train)](<https://www.sota2.com/research/sota/multi-hop-question-answering-on-hotpotqa-train.md>): Multi-hop Question Answering; HotpotQA (train); Accuracy: 0.306.

- [Single-Hop Question Answering on NQ (train)](<https://www.sota2.com/research/sota/single-hop-question-answering-on-nq-train.md>): Single-hop Question Answering; NQ (train); Accuracy: 32.8.

- [Single-Hop Question Answering on PopQA](<https://www.sota2.com/research/sota/single-hop-question-answering-on-popqa.md>): Single-hop Question Answering; PopQA; EM: 36.8.

- [Multi-hop Question Answering on 2Wiki](<https://www.sota2.com/research/sota/multi-hop-question-answering-on-2wiki.md>): Multi-hop Question Answering; 2Wiki; Exact Match: not recorded.

- [Multi-hop Question Answering on Musique](<https://www.sota2.com/research/sota/multi-hop-question-answering-on-musique.md>): Multi-hop Question Answering; MuSiQue; EM: not recorded.

- [Single-hop Question Answering on TriviaQA](<https://www.sota2.com/research/sota/single-hop-question-answering-on-triviaqa.md>): Single-hop Question Answering; TriviaQA; EM: not recorded.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2601.03261>)

[PDF](<https://arxiv.org/pdf/2601.03261.pdf>)
