StructSurvey: Structured Agentic Retrieval for Automatic Survey Paper Generation

Paolo Pedinotti, Enrico Santus

Keywords: automated survey generation, structured retrieval, multi-agent systems, retrieval-augmented generation, graph-based retrieval, structured information extraction, domain graphs, hierarchical planning, long-form scientific summarization, large language models, LLM-based writing, survey paper synthesis, reference-grounded evaluation, ACL surveys, ROUGE evaluation, LLM-as-a-Judge, scientific document organization, knowledge graphs, taxonomy induction

TL;DR: The paper introduces StructSurvey, a hierarchical multi-agent system for automatic survey paper generation that replaces unstructured retrieval with structured, graph-based retrieval of entities, relations, and taxonomies.

Abstract: The explosive growth of scientific publications has made it increasingly difficult to track and synthesize developments across research fields. While Large Language Models (LLMs) offer a path toward automated survey paper generation, existing methods rely on unstructured retrieval, forcing models to infer how concepts, methods, and research directions relate based on raw text alone. We introduce STRUCTSURVEY, a hierarchical, multi-agent framework for automated survey paper generation that shifts structural reasoning from generation to retrieval by dynamically constructing graph-based representations of entities, relations, and topical taxonomies. We evaluate STRUCTSURVEY on a new reference-grounded benchmark of ACL survey papers, which we construct to sup- port reproducible evaluation of long-form scientific summarization. STRUCTSURVEY achieves consistent improvements over embedding-only retrieval baselines, including +2.9 ROUGE-1 recall and +1.0 ROUGE-2 recall on average, with no loss in precision. Beyond lexical overlap, STRUCTSURVEY also improves LLM-as-a-Judge ratings for generated surveys along dimensions of logical structure, depth, and synthesis under a principled, robustness-oriented evaluation protocol, demonstrating that explicit structural retrieval enables surveys that more closely match human-written organization and reasoning.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 5