发表机构
National University of Singapore; TCG CREST; Indian Institute of Technology Kharagpur; Singapore Institute of Technology; Indian Institute of Technology Bombay(新加坡国立大学; TCG CREST; 印度理工学院卡拉格普尔分校; 新加坡理工学院; 印度理工学院孟买分校)
机构由 AI 辅助整理,请以论文原文为准。AI 中文总结
Crase是一种有界可检查的学术搜索方法,通过1.5跳引用邻域扩展等操作,在50万篇arXiv论文的基准上,召回率@50最高达同类深度研究智能体的3倍且成本仅为其三分之一。
AI 中文摘要
我们提出了Crase,一种用于学术搜索的、有界且可检查的深度研究智能体替代方案。它不采用开放式搜索循环,而是仅向搜索引擎查询一次以获取种子论文,沿其1.5跳引用邻域扩展,修剪那些主张缺乏蕴含支持的引用边,并使用感知时效性的随机游走对剩余论文进行排名。这使得候选集、保留每篇论文的原因以及停止条件在推理前就明确且固定。在LitSearch基准及另一个基于50万篇论文的arXiv语料库的基准测试中,Crase的召回率@50最高达到基于专有模型构建的深度研究智能体的3倍,而成本仅约为其三分之一。
英文摘要
We present Crase, a bounded and inspectable alternative to deep research agents for scholarly search. Instead of an open-ended search loop, Crase queries a search engine once for seed papers, expands them along their 1.5-hop citation neighborhood, prunes citation edges whose claims lack entailment support, and ranks the remaining papers with a recency-aware random walk. This makes the candidate set, the reason each paper is kept, and the stopping condition explicit and fixed before inference. On LitSearch and one further benchmarks over a 500K-paper arXiv corpus, Crase outperforms deep research agents built on proprietary models by up to 3$\times$ recall@50 at roughly a third of the cost.