Elamur Research⌕
☰
home / papers / 2026 / StructSurvey: Structured Agentic Retrieval for Automated Survey Paper Generation

StructSurvey: Structured Agentic Retrieval for Automated Survey Paper Generation

Paper · 2026 · Paolo Pedinotti and Enrico Santus

Summary

Automated survey writing with LLMs usually relies on retrieval that hands the model plain passages. The model then has to work out for itself how methods, tasks, datasets and research threads connect while it is writing. StructSurvey changes this by doing that structural work during retrieval. It is a hierarchical multi-agent system in which a planner attaches a typed query function to each outline node. A query function returns either ordinary vector-retrieved abstracts or an LLM-extracted graph fragment of entities, relations and small induced taxonomies. These fragments are merged into a growing domain graph that writer agents consult when drafting subsections.

To evaluate the approach, the authors built a reference-grounded benchmark, the ACL Survey Dataset. Each survey's retrieval pool is limited to its cited papers, and only abstracts are used. Against embedding-only systems such as AutoSurvey and SurveyForge, StructSurvey raises ROUGE recall without lowering precision. LLM judges also rate its outputs higher on organization, depth and synthesis. The work shows that explicit structure helps long-form scientific summarization.

Key points

  • Benchmark: the ACL Survey Dataset contains 33 ACL survey papers from 2018–2025, each with full text and publicly available references.
  • ROUGE gains: compared with embedding-only baselines, StructSurvey improves ROUGE-1 recall by +2.9 and ROUGE-2 recall by +1.0 on average, while precision stays comparable.
  • Query functions: there are three graph query types: single_entity_search, related_entity_search and pair_of_related_entities_search.
    • related_entity_search correlates most strongly with ROUGE improvement over SurveyForge (ρ = 0.380).
    • The other two correlate less (0.197 and 0.114).
    • This holds even though related_entity_search is used less often than single_entity_search (3.1 vs. 7.1 uses per paper).
  • LLM-as-a-Judge results: StructSurvey scores better on logical structure and depth, and significantly better on synthesis. Critical analysis remains weak across all systems.
  • Limitations:
    • Structured queries are costly, requiring about D + 1 + ⌈E/b⌉ LLM calls each.
    • The evaluation covers only an abstract-level, reference-grounded setting, not open-corpus retrieval.

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

 Institutions

🏢BloombergNew York · United States7 works · 1,559 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksInformation Systems · 1,437 works

🧭 Topics 1Research subjects and languages the work is about.Information retrieval & search · 473 works

Cite

Cite as Paolo Pedinotti and Enrico Santus. 2026. StructSurvey: Structured Agentic Retrieval for Automated Survey Paper Generation. Proceedings of the First Workshop on Structured Understanding, Retrieval, and Generation in the LLM Era (SURGeLLM 2026).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.surgellm-1.10

@inproceedings{pedinotti2026structsurvey,
  title = {{StructSurvey: Structured Agentic Retrieval for Automated Survey Paper Generation}},
  author = {Paolo Pedinotti and Enrico Santus},
  booktitle = {Proceedings of the First Workshop on Structured Understanding, Retrieval, and Generation in the LLM Era (SURGeLLM 2026)},
  year = {2026},
  doi = {10.18653/v1/2026.surgellm-1.10},
  url = {https://doi.org/10.18653/v1/2026.surgellm-1.10}
}

References

7 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Joachim Baumann, Paul Röttger, Aleksandra Urman, Albert Wendsjö, Flor Miriam Plaza del Arco, Johannes B. Gruber, and Dirk Hovy. 2025. Large language model hacking: Quantifying the hidden risks of using llms for text annotation. Preprint, arXiv:2509.08825. — [arXiv:2509.08825](https://arxiv.org/abs/2509.08825) find
  2. Matthijs Douze, Alexandr Guzhva, Chengqi Deng, Jeff Johnson, Gergely Szilvasy, Pierre-Emmanuel Mazaré, Maria Lomeli, Lucas Hosseini, and Hervé Jégou. 2025. The faiss library. IEEE Transactions on Big Data.
  3. Darren Edge, Ha Trinh, Newman Cheng, Joshua Bradley, Alex Chao, Apurva Mody, Steven Truitt, Dasha Metropolitansky, Robert Osazuwa Ness, and Jonathan Larson. 2024. From local to global: A graph rag approach to query-focused summarization. arXiv preprint arXiv:2404.16130. GraphRAG constructs an LLM-de… find
  4. Jiawei Gu, Xuhui Jiang, Zhichao Shi, Hexiang Tan, Xuehao Zhai, Chengjin Xu, Wei Li, Yinghan Shen, Shengjie Ma, Honghao Liu, and 1 others. 2024. A survey on llm-as-a-judge. The Innovation. find
  5. Haoyu Han, Yu Wang, Harry Shomer, Kai Guo, Jiayuan Ding, Yongjia Lei, Mahantesh M. Halappanavar, Ryan A. Rossi, Subhabrata Mukherjee, Xianfeng Tang, Qianru He, Zhigang Hua, Bo Long, Tong Zhao, Neil Shah, Amin Javari, Yinglong Xia, and Jiliang Tang. 2024. Retrieval-augmented generation with graphs (g… find
  6. Pengcheng Jiang, Siru Ouyang, Yizhu Jiao, Ming Zhong, Runchu Tian, and Jiawei Han. 2025. Retrieval and structuring augmented generation with large language models. In Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.2, KDD ’25, page 6032–6042. ACM. find
  7. Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, Sebastian Riedel, and Douwe Kiela. 2020. Retrieval-augmented generation for knowledge-intensive nlp tasks. In Proceedings of the 34th Internation…

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/pedinotti-2026-structsurvey-structured-agentic-retrieval-automated/