SurveyForge - An AI-powered framework for generating automated academic reviews

What is SurveyForge?

SurveyForge is an innovative framework developed by the Shanghai AI Lab in collaboration with Fudan University, Shanghai Jiao Tong University, and other institutions. It's used to automatically generate high-quality academic reviews. The framework is based on a two-stage design: outline generation and content generation. In the outline generation stage, a heuristic learning mechanism driven by a dual-database collaboration (a research paper database and a review outline database) generates a well-structured and logically clear review framework. In the content generation stage, the Scholar Navigation Agent (SANA) uses a memory module and a time-aware reordering engine to accurately retrieve literature and generate high-quality content. SurveyForge's parallel generation and coordination mechanism can efficiently generate coherent long documents; generating a review of approximately 64k tokens takes only 10 minutes and costs less than $0.50. The team has also built SurveyBench, a multi-dimensional evaluation benchmark used to objectively assess the quality of reviews.

SurveyForge's main functions

• Efficiently generate reviewsSurveyForge can generate a summary of approximately 64k tokens in a short time, with the entire process taking only 10 minutes and costing less than $0.50.
• High-quality outline generationBased on a heuristic learning mechanism driven by dual databases, it generates a well-structured and logically clear review outline, approaching the level of human writing.
• Precise literature retrievalScholar Navigation Agent (SANA) can accurately retrieve high-quality literature related to the topic, avoid citing irrelevant literature, and improve the accuracy and impact of the literature.
• Content optimization and coordinationThe jiyu parallel generation and coordination mechanism allows each chapter to be generated independently, and then the refining stage is used to eliminate duplication and streamline the logic to form a coherent whole.

SurveyForge's technical principles

• Outline generation stage:
  • Dual database collaborative driveThe project constructs a research paper database (approximately 600,000 arXiv papers in the field of computer science) and a review outline database (a hierarchical outline structure of approximately 20,000 review articles). Based on cross-database knowledge fusion, it retrieves topic-related papers and existing review outlines, learning expert-level structured patterns.
  • Recursive construction strategyFirst, a primary outline reflecting the overall logic is generated. Then, the secondary structure is further refined for each chapter in conjunction with relevant literature, achieving structured learning from coarse to fine and from the whole to the parts.
• Content generation stage:
  • Scholar Navigation Agent (SANA):
    • Subquery memory moduleThe literature collection retrieved during the outline generation stage is used as a memory context, combined with the original query, to ensure that the query decomposition process always revolves around the core topic and avoids semantic deviation.
    • Retrieval Memory ModuleThe entire outline-related literature is treated as a global memory, and the most relevant literature is retrieved for each subquery based on embedding similarity, thereby improving retrieval accuracy and semantic consistency.
    • Time-aware reordering engineThe retrieved literature is grouped by publication date, and then top-k filtered by citation count within each group to balance classic authoritative literature with cutting-edge and emerging research.
  • Parallel generation and coordinationEach chapter generates content independently, and a shared memory system ensures that the content revolves around a unified thematic framework. The refining stage eliminates repetition, clarifies logic, and forms a coherent whole.

SurveyForge project address

• GitHub repositoryhttps://github.com/Alpha-Innovator/SurveyForge
• HuggingFace model libraryhttps://huggingface.co/datasets/U4R/SurveyBench
• arXiv technical paper: https://arxiv.org/pdf/2503.04629

Application scenarios of SurveyForge

• academic researchIt helps newcomers to the field, interdisciplinary researchers, and seasoned scholars quickly gain a comprehensive understanding of the field and improve the efficiency of literature review.
• EducationIt assists teachers in designing courses and students in learning, enabling them to quickly master domain knowledge.
• industryIt helps companies conduct technology research and industry analysis, providing reference for R&D decisions.
• Policy makingIt provides scientific evidence for policymakers and assists in technology assessment and policy planning.
• Personal learning and researchIt helps individuals quickly understand the latest developments in their areas of interest and assists in thesis writing.