What is SurveyX
SurveyX is a system for automatically generating academic surveys based on large language models (LLMs), jointly launched by Renmin University of China, the University of Sydney, and Northeastern University of China. Based on the paper title and keywords provided by the user, it can quickly generate high-quality, domain-specific academic surveys orpapers. The SurveyX system is based on advanced language model technology and combines data processing and literature retrieval capabilities to help users save time and effort in writing academic surveys. SurveyX decomposes the survey generation process into a preparation stage and a generation stage, addressing issues in traditional methods such as context window limitations, outdated knowledge, and the lack of a systematic evaluation framework. SurveyX outperforms existing methods in content quality, citation quality, and literature relevance, approaching the level of human experts, and provides strong support for efficiently generating high-quality academic surveys.
SurveyX's main features
- Automated generation of academic surveys: Users provide a paper title and related keywords, and the system automatically generates a high-quality academic survey or research paper.
- Customized content generation: Users can specify the literature retrieval scope based on keywords according to their own research needs, and generate survey content in a specific field.
- Efficient literature retrieval and integration: Retrieve relevant literature based on keywords, integrate information from the literature, and generate comprehensive and structured survey content.
- Support for multiple academic fields: The system is applicable to a variety of disciplinary fields, including but not limited to artificial intelligence, natural language processing, computer science, medicine, physics, etc.
SurveyX's technical principles
- Keyword expansion and literature retrieval: Based on a keyword expansion algorithm, through semantic clustering and keyword extraction, the retrieval keyword pool is gradually expanded to ensure the comprehensiveness of retrieval. Combined with a two-step filtering method, an embedding model is used for coarse-grained filtering, and LLMs are used for fine-grained filtering, ensuring that the literature is highly relevant to the topic.
- Literature preprocessing: Extract key information from the literature and construct it into an attribute tree, significantly improving the information density of the literature and the utilization of the LLMs' context window. Different types of literature (such as method papers, theoretical papers, etc.) use different attribute tree templates to ensure the targeting and accuracy of information extraction.
- Intelligent outline generation: Generate hints based on AttributeTree to assist LLMs in generating second-level outlines. Based on the 'separate-recombine' steps, redundancy is eliminated, and the logical structure of the outline is optimized to ensure the organization and coherence of the survey.
- Content generation and optimization: Based on RAG technology, combined with the retrieved literature materials, optimize the citation quality and accuracy of the generated content. During the generation process, LLMs can view the content of other subsections to ensure the consistency of the generated content.
- Multimodal presentation and post-processing: Based on information extraction templates and generation templates, extract the necessary information from the literature to generate figures and tables, enriching the presentation of the survey. Use multimodal LLMs to retrieve figures and tables that support the content, further improving the readability of the survey and the effectiveness of information communication. Refine and polish the generated draft to improve text fluency, logic, and academic quality.
SurveyX's project address
- Project official website：surveyx.cn
- GitHub repository：https://github.com/IAAR-Shanghai/SurveyX
- HuggingFace model library：https://huggingface.co/papers/2502.14776
- arXiv technical paper：https://arxiv.org/pdf/2502.14776
SurveyX's application scenarios
- Academic research: Quickly generate high-quality literature reviews, helping researchers grasp the current state of the field and research directions.
- Interdisciplinary Research: Integrate multidisciplinary literature and promote interdisciplinary knowledge integration.
- Dynamic Updates: Retrieve the latest literature in real time and generate dynamic reviews to help track cutting-edge research.
- Teaching Assistance: Help students learn review writing and improve their academic writing skills.
- Industry Analysis: Generate technology reviews and industry reports to provide decision-making references for enterprises and institutions.