中文

BioDisco: 基于双模证据、迭代反馈和时序评估的多智能体假设生成

人工智能 2025-11-25 v2 信息检索 应用统计

摘要

识别新假设是科学研究的关键环节,但由于信息量庞大且复杂,这一过程容易被淹没。现有自动化方法常常难以生成新颖且以证据为依据的假设,缺乏健壮的迭代细化机制,且很少进行严格的时序评估以评估未来发现潜力。为此,我们提出了 BioDisco,一个多智能体框架,基于语言模型推理,运用双模证据系统(生物医学知识图谱和自动文献检索)实现依据化的新颖性,集成内部评分与反馈循环进行迭代细化,并通过首创性时序与人类评估以及Bradley-Terry配对比较模型提供统计学依据的评估。我们的评估表明,与被剥离配置和通用生物医学智能体相比,BioDisco在新颖性和显著性方面表现出优势。 designed for flexibility and modularity, BioDisco allows seamless integration of custom language models or knowledge graphs, and can be run with just a few lines of code.

关键词

引用

@article{arxiv.2508.01285,
  title  = {BioDisco: Multi-agent hypothesis generation with dual-mode evidence, iterative feedback and temporal evaluation},
  author = {Yujing Ke and Kevin George and Kathan Pandya and David Blumenthal and Maximilian Sprang and Gerrit Großmann and Sebastian Vollmer and David Antony Selby},
  journal= {arXiv preprint arXiv:2508.01285},
  year   = {2025}
}

备注

12 pages main content, 31 including appendices. 8 figures