Structurally-bounded Agentic Graph Exploration for Evidence-Grounded Scholarly DeepSearch

2026-08-25

AI Summary

该研究提出Crase方法，通过一次性查询、扩展引用邻域、修剪无支持的引用并排名论文来改进学术搜索，相比深度研究代理具有更高的召回率和更低的成本。

Abstract

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.

Problem

• scholarly search
• deep research agents
• inspectable
• bounded
• citation neighborhood

Innovation

• structurally-bounded
• inspectable
• 1.5-hop citation neighborhood
• recency-aware random walk
• cost-effective