Skip to content

AutoSurvey

external · status: dormant · focus: literature · discipline: general · started: 2024

Project page: https://github.com/AutoSurveys/AutoSurvey

Source: projects/landscape/autosurvey.yml

Positioning

A NeurIPS 2024 framework (arXiv:2406.10252) for automatically generating comprehensive literature surveys from a topic and a paper database. Demonstrated on survey lengths of 8k–64k tokens with reported citation-quality and content-quality scores. Sits squarely in the literature-synthesis stage of the RISE diagram.

Distinctive contribution

Among the first systems to treat long-form survey writing (not short-form QA or summarization) as the target task, with explicit evaluation of citation quality at scale. Ships with a 530K-abstract arXiv-CS database used in the published experiments.

Evaluation scores

Dimension Score (0–3) Note
Lifecycle coverage 1 Three stages clustered around literature synthesis and drafting.
Autonomy level 3 Runs end-to-end from a topic to a survey; no per-step approval.
Architectural transparency 2 NeurIPS 2024 paper documents the framework; code and database public; prompts visible.
Inputs supported 1 Topic input only; database is fixed (CS-arXiv abstracts in the public release).
Outputs / reproducibility 2 Code + database + commands published for paper experiments.
Internal evaluation 3 Systematic evaluation across multiple survey lengths in the NeurIPS paper.
Openness 1 No license declared in repository metadata — defaults to all rights reserved; database access via OneDrive link from maintainers.
Maturity / traction 1 468 stars; activity slowed sharply after the NeurIPS publication (last push 2025-02).
Cross-family policy 0 Single LLM per run.
Runtime assurance 1 Citation-quality and content-quality scoring in NeurIPS paper; no in-pipeline claim audit harness.
Cross-platform portability 1 Code + paper-DB available; single back-end.

Scored on 2026-05-18. See the evaluation rubric.

Tags

Pipeline stages: literature-discovery literature-synthesis paper-drafting

Architectural features: multi-agent rag-knowledge-base iterative-loop

Inputs: survey-topic

Outputs: long-form-survey citations

Data sources: arxiv-cs-abstracts

Knowledge sources: arxiv

Limitations

  • No declared open-source license — reuse rights are uncertain.
  • Database limited to CS-arXiv abstracts in the public release; full-text version requires contacting authors.
  • Last commit ~2025-02; appears semi-maintained.

Papers describing this project

  • AutoSurvey: Large Language Models Can Automatically Write Surveys — Wang, Y., Guo, Q., Yao, W., Zhang, H., Zhang, X., Wu, Z., et al. (2024). NeurIPS 2024. arXiv:2406.10252