SurveyGO - An open-source AI-powered thesis writing tool developed in collaboration with Tsinghua University and Mianbi Intelligence.

What is SurveyGO?

SurveyGO is an open-source AI-powered paper writing tool developed by Tsinghua University in collaboration with the Facewall AI team. It supports the rapid generation of well-structured and content-rich review articles based on user-inputted topics and descriptions, helping researchers and students efficiently complete their literature reviews. SurveyGO's core technology, LLMxMapReduce-V2, draws inspiration from convolutional neural networks, aggregating multiple references using a text convolution algorithm to avoid information loss issues common in traditional methods. The generated articles are logically rigorous, reliably cited, and rich in length and detail. Users simply fill in information and submit their topic selection; the system generates high-quality review documents in the background, significantly improving research writing efficiency.

Main functions of SurveyGO

• Quickly generate review articlesAfter users input the topic and description of their paper, the system can quickly generate well-structured and rich reviews, supporting the output of long articles of tens of thousands of words.
• User InteractionBased on the like and comment functions, users provide feedback on their writing needs and the generated articles, and the system optimizes content generation based on user feedback.
• Multilingual supportIt offers options for generating reviews in both Chinese and English to meet the needs of different users.
• User-friendly interfaceIt is easy to use, supports both normal and professional modes, and makes it convenient for users to submit topic selections and view results.

The technical principles of SurveyGO

• LLMxMapReduce-V2 technologyThis model aggregates multiple references using a text convolution algorithm. Similar to the application of Convolutional Neural Networks (CNNs) in image processing, it progressively abstracts local information into a high-level global representation. The model first focuses on some cited articles (local information), and then, based on multi-layer convolution operations, integrates this local information into more comprehensive structured information, such as the paragraph structure and topic of the article. An information entropy estimation module is introduced to guide the convolution process, ensuring that the information content of the results is continuously improved during scaling during testing.
• Resolving context length limitationsTraditional Search Augmentation (RAG) methods generate content based on the most relevant text fragments to the query, easily overlooking important content that, while related to the topic, cannot be directly matched semantically. Our approach, however, aggregates multiple references instead of simple searches, fully utilizing information from all references to avoid information loss.
• High-quality content generationBased on multi-layer convolutional operations, key information is extracted from the literature and integrated into a structured output, ensuring the logic and coherence of the generated article. Relevant literature and materials are automatically cited to ensure the credibility and professionalism of the generated content.
• Evaluation benchmarkSurveyEval is the first scalable evaluation benchmark in computer science that combines surveys with complete bibliographies to scientifically assess the quality of generated articles and create a high-quality benchmark for survey writing.

How to use SurveyGO

• Visit the websiteAccessing SurveyGOOfficial websiteFollow the prompts to complete the registration and login process.
• Select mode:
  • Normal modeSuitable for beginners or users who do not have high requirements for the generated content.
  • Professional modeSuitable for users who have higher requirements for the content of the review.
• Fill in the information:
  • Paper titleEnter the topic you want to generate the review on.
  • Paper DescriptionBriefly describe your research direction or the key areas you need to focus on.
  • Language selectionSelect the language of the generated article (Chinese or English).
• Submit topic selectionAfter you fill in the information and submit it, the system will automatically start generating a review article.
• View the generated resultsOnce generated, view the generated article.

SurveyGO's project address

• Project official website:https://surveygo.thunlp.org/
• GitHub repository:https://github.com/thunlp/LLMxMapReduce
• HuggingFace model library:https://huggingface.co/datasets/R0k1e/SurveyEval
• arXiv technical paper:https://arxiv.org/pdf/2504.05732

Application scenarios of SurveyGO

• Scientific paper writingIt helps researchers quickly generate literature review sections, saving time spent searching for information and organizing ideas.
• Preparation of academic reportsWhen preparing academic reports, it can quickly provide background information and research progress on relevant topics, helping users to quickly build a report framework.
• Course paper writingIt is a powerful tool for students to write the review section of course papers, providing a clear logical structure and rich reference content.
• Industry Research and AnalysisIt is used to generate research reports for specific industries, analyze market trends and technological developments, and provide references for business decisions.
• Learning and Knowledge ExpansionQuickly gain insights into core viewpoints and research trends in new fields, broadening your knowledge base.