Elamur Research⌕
☰
home / papers / 2026 / NEST: Nested Evidence Survival for Retrieval

NEST: Nested Evidence Survival for Retrieval

Paper · 2026 · Akshay Verma, Siddharth Pillai, Prateek Sircar et al.

Summary

Retrieval-augmented generation tends to break down when the corpus is flooded with distractors. In that setting, useful passages are sparse, and fixed top-k or static chunking can discard them before their relevance becomes clear. Adaptive alternatives such as DeepRAG or AutoRAG address this with learned or iterative control, but they need extra training or repeated retrieval, which raises cost and latency. The authors, from Amazon, propose NEST, a training-free pipeline with two stages. The first stage scores candidates at document, section and chunk level within one retrieval pass and keeps everything found at any level, so that weakly phrased or multi-hop evidence can be recovered once broader context supports it. The second stage, NR-MRR, ranks candidates by their reciprocal ranks across these nested scopes and keeps the top-k, which reduces redundancy. The method is tested on three benchmarks: WebQuestions, the distractor configuration of HotpotQA, and InternalQA, a dataset the authors keep private. NEST beats DeepRAG on accuracy and recall at latency close to Vanilla RAG. The work suggests that restructuring retrieval itself can be more cost-effective than adding controllers or larger models.

Key points

  • Benchmarks: WebQuestions, HotpotQA (distractor) and InternalQA, an in-house benchmark padded with 50M Common Crawl distractor documents.
  • Baselines: Vanilla RAG (k=20), Iterative RAG, AutoRAG and DeepRAG. Metrics are EM, F1, Recall@50, a redundancy ratio and latency.
  • WebQuestions results:
    • Against DeepRAG, NEST reaches 62.8 EM vs 60.4 and 77.3 F1 vs 75.2.
    • Recall@50 rises from 78.1 to 84.9.
    • Redundancy falls to 0.47, against 0.55 for DeepRAG and 0.68 for Vanilla.
  • Latency: about 836 ms for NEST, vs 820 ms for Vanilla RAG and 1,650 ms for DeepRAG. The paper puts the overhead at roughly 12–18 ms, of which NR-MRR accounts for under 2 ms.
  • Ablations on WebQuestions:
    • Without nested scopes, Recall@50 drops to 75.8 and EM to 54.1.
    • Without NR-MRR, recall stays at 84.3, but redundancy rises to 0.63 and EM falls to 56.1.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Fields, topics and concepts

📚 Shelves 6The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksComputer Vision and Pattern Recognition · 3,486 worksInformation Systems · 1,730 works

🧭 Topics 1Research subjects and languages the work is about.Information retrieval & search · 552 works

Cite

Cite as Akshay Verma, Siddharth Pillai, Prateek Sircar and Deepak Gupta. 2026. NEST: Nested Evidence Survival for Retrieval. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 6: Industry Track).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.acl-industry.35

@inproceedings{verma2026nest,
  title = {{NEST: Nested Evidence Survival for Retrieval}},
  author = {Akshay Verma, Siddharth Pillai, Prateek Sircar and Deepak Gupta},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 6: Industry Track)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-industry.35},
  url = {https://doi.org/10.18653/v1/2026.acl-industry.35}
}

References

22 references, 9 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Ali Behrouz, Meisam Razaviyayn, Peilin Zhong, and Vahab Mirrokni. 2025. Nested learning: The illusion of deep learning architectures. In Advances in Neural Information Processing Systems (NeurIPS) 2025. Poster. find
  2. Jonathan Berant, Andrew Chou, Roy Frostig, and Percy Liang. 2013. Semantic parsing on Freebase from question-answer pairs. In Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 1533–1544, Seattle, Washington, USA. Association for Computational Linguistics.
  3. Chi-Min Chan, Chunpu Xu, Ruibin Yuan, Hongyin Luo, Wei Xue, Yi-Ting Guo, and Jie Fu. 2024. Rq-rag: Learning to refine queries for retrieval augmented generation. ArXiv, abs/2404.00610.
  4. Patrick Lewis, Ethan Perez, Aleksandara Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Kuttler, Mike Lewis, Wen tau Yih, Tim Rocktäschel, Sebastian Riedel, and Douwe Kiela. 2020. Retrieval-augmented generation for knowledge-intensive nlp tasks. ArXiv, abs/2005.11401.
  5. Ruiyang Ren, Yingqi Qu, Jing Liu, Wayne Xin Zhao, QiaoQiao She, Hua Wu, Haifeng Wang, and Ji-Rong Wen. 2021. RocketQAv2: A joint training method for dense passage retrieval and passage re-ranking. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 2825–2…
  6. Ishneet Sukhvinder Singh, Ritvik Aggarwal, Ibrahim Allahverdiyev, Muhammad Taha, Aslihan Akalin, Kevin Zhu, and Sean O’Brien. 2024. Chunkrag: Novel llm-chunk filtering method for rag systems. ArXiv, abs/2410.19572. — [arXiv:2410.19572](https://arxiv.org/abs/2410.19572) find
  7. Yixiong Fang, Tianran Sun, Yuling Shi, and Xiaodong Gu. 2025. Attentionrag: Attention-guided context pruning in retrieval-augmented generation. find
  8. Sambeet Tiady, Anirban Majumder, and Deepak Gupta. 2023. Prodigy: Product design guidance at scale. In CIKM, pages 4836–4842. find
  9. Xinyan Guan, Jiali Zeng, Fandong Meng, Chunlei Xin, Yaojie Lu, Hongyu Lin, Xianpei Han, Le Sun, and Jie Zhou. 2025. Deeprag: Thinking to retrieve step by step for large language models. find
  10. Gautier Izacard, Mathilde Caron, Lucas Hosseini, Sebastian Riedel, Piotr Bojanowski, Armand Joulin, and Edouard Grave. 2021. Unsupervised dense information retrieval with contrastive learning.
  11. Trans. Mach. Learn. Res., 2022. find
  12. Gautier Izacard, Patrick Lewis, Maria Lomeli, Lucas Hosseini, Fabio Petroni, Timo Schick, Jane A. Yu, Armand Joulin, Sebastian Riedel, and Edouard Grave. 2022. Few-shot learning with retrieval augmented language models. J. Mach. Learn. Res., 24:251:1– 251:43. find
  13. Arihant Jain, Purav Aggarwal, and Anoop Saladi. 2025. AutoChunker: Structured text chunking and its evaluation. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 6: Industry Track), pages 983–995, Vienna, Austria. Association for Computational Linguis… find
  14. Shuguang Jiao, Xinyu Xiao, Yunfan Wei, Shuhan Qi, Chengkai Huang, Quan Z. Michael Sheng, and Lina Yao. 2026. Prunerag: Confidence-guided query decomposition trees for efficient retrieval-augmented generation. find
  15. Anant Khandelwal, Happy Mittal, Shreyas Sunil Kulkarni, and Deepak Gupta. 2023. Large scale generative multimodal attribute extraction for e-commerce attributes. CoRR, abs/2306.00379. find
  16. Dongkyu Kim, Byoungwook Kim, Donggeon Han, and Matouvs Eibich. 2024. Autorag: Automated framework for optimization of retrieval augmented generation pipeline. ArXiv, abs/2410.20878. — [arXiv:2410.20878](https://arxiv.org/abs/2410.20878) find
  17. Harsh Trivedi, Niranjan Balasubramanian, Tushar Khot, and Ashish Sabharwal. 2023. Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), p…
  18. Akshay Verma, Swapnil Gupta, Deepak Gupta, Prateek Sircar, and Siddharth Pillai. 2026a. SELENE: Selective and evidence-weighted LLM debating for efficient and reliable reasoning. In Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 5:…
  19. Akshay Verma, Swapnil Gupta, Siddharth Pillai, Prateek Sircar, and Deepak Gupta. 2026b. ReflectiveRAG: Rethinking adaptivity in retrieval-augmented generation. In Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 5: Industry Track), p…
  20. Vinay Kumar Verma, Shreyas Sunil Kulkarni, Happy Mittal, and Deepak Gupta. 2025. Moemoe: Question guided dense and scalable sparse mixture-of-expert for multi-source multi-modal answering. arXiv preprint arXiv:2503.06296. find
  21. Zhilin Yang, Peng Qi, Saizheng Zhang, Yoshua Bengio, William W. Cohen, Ruslan Salakhutdinov, and Christopher D. Manning. 2018. Hotpotqa: A dataset for diverse, explainable multi-hop question answering. In Conference on Empirical Methods in Natural Language Processing.
  22. Siyuan Zhu, Chengdong Xu, Kaiqiang Ke, and Chao Yu. 2026. Context-picker: Dynamic context selection using multi-stage reinforcement learning. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/verma-2026-nest-nested-evidence-survival-retrieval/