SurveyForge: An AI Tool for Automatically Writing Survey Papers 79

小编 · 2025-06-13

SurveyForge is a tool developed by Shanghai AI Lab, dedicated to automatically generating survey papers. It studies the outline structures of human-written papers and related literature to produce reasonable outlines, then uses high-quality academic papers to automatically generate and refine content. The authors also created the SurveyBench benchmark, containing 100 human-written survey papers, to evaluate AI-generated papers from multiple perspectives. This addresses common issues in AI-generated papers such as unreasonable structure and inaccurate references.

SurveyForge: An AI Tool for Automatically Writing Survey Papers

What is SurveyForge?

SurveyForge is a tool developed by Shanghai ai Lab, specifically designed to automatically generate survey papers. It generates reasonable outlines by studying the structure of human-written paper outlines and related literature, and uses high-quality academic papers to automatically generate and optimize content. In addition, the authors created the SurveyBench benchmark, which contains 100 human-written survey papers to evaluate the quality of AI-generated papers from multiple perspectives. This addresses common issues in traditional AI-generated papers, such as unreasonable structure and inaccurate references.

SurveyForge: An AI tool for automatically writing survey papers.webp

Core Features

Outline Generation: SurveyForge studies human-written paper outlines and related literature to generate clear, reasonable structures.
Content Generation: It retrieves high-quality literature and combines a memory mechanism to generate and optimize paper content.
Multi-dimensional Evaluation: The SurveyBench benchmark contains 100 human-written survey papers to evaluate AI-generated surveys from three aspects: reference quality, outline quality, and content quality.

Technical Details

Outline Generation: Creates outlines by analyzing the structure of human-written papers and related domain literature.
Content Generation: Uses high-quality literature to generate and optimize paper content.
Evaluation Benchmark: SurveyBench contains 100 human-written papers across 10 different topics, ensuring topical consistency and academic rigor.

Advantages

Structure and Logic: The quality of generated outlines is close to human level, solving the structural imbalance problem in traditional AI methods.
References: The coverage rate of core literature nearly doubled, with more accurate citations.
Content Quality: The generated content is more coherent and comprehensive, at very low cost—generating a survey of approximately 64k tokens costs less than $0.5, and the entire process takes only 10 minutes.
Evaluation Standard: SurveyBench provides objective evaluation criteria, making it easy to compare the performance of different methods.

How to Use

Generate Papers: Due to API call rate limits, users need to request paper generation via email or by submitting an issue in the repository. Generation time is approximately 10 minutes.
Evaluate Papers: Use SurveyBench for evaluation, which can be done by running cd SurveyBench && python test.py --is_human_eval from the command line.