迈向自我进化的智能文献检索系统
针对传统检索无法理解复杂意图、而前沿大语言模型成本高且存在幻觉的问题，研究团队提出了自我进化的智能文献检索系统PaSaMaster。该系统通过迭代式意图分析、检索与排序，将文献检索转变为动态演进的过程，并采用三项关键设计：利用排序证据揭示信息缺口以优化搜索；将检索定义为意图-论文相关性排序任务，从根本上杜绝虚假文献；通过分离规划与检索来提升效率，仅用大模型理解意图，而将大规模检索与评分交由轻量模型处理。在涵盖38个学科的基准测试中，该系统将传统关键词检索的F1分数提升15.6倍，完全消除了文献幻觉，且性能超越GPT-5.2达30%，计算成本仅为后者的1%。
学术文献检索一直被关键词和LLM幻觉两头堵，这个系统用规划与检索分离做到了零幻觉，F1暴涨15.6倍，比GPT-5.2强30%却只花1%算力，做科研的可以马上跑起来。
Authors:Yuwen Du, Tian Jin, Jing Kang, Xianghe Pang, Jingyi Chai, Tingjia Miao, Fenyi Liu, WenHao Wang, Sikai Yao, Yuzhi Zhang, Siheng Chen
Abstract:Scientific literature retrieval must understand complex search intents while preserving source authenticity. Traditional keyword and embedding-based systems return authentic sources but miss nuanced intents, whereas large language models capture richer intents but may fabricate citations. We introduce PaSaMaster, a Recursive Self-Evolving agentic literature retrieval system that iteratively analyzes intent, retrieves verified papers and ranks them with evidence-grounded relevance scores. PaSaMaster combines self-evolving retrieval that refines search intent from ranked evidence over time, hallucination-free ranking over verified papers rather than generated citations, and cost-efficient planning--retrieval separation that reserves frontier LLMs for intent understanding while delegating retrieval and scoring to lightweight models and customized corpora. Across 38 disciplines in PaSaMaster-Bench, PaSaMaster achieves a 16.5$\times$ higher F1-score than Google Scholar and a 37.8\% higher F1-score than GPT-5.2 at about 1\% of the cost, while reducing source hallucination from 32.66\% in generative LLMs to zero: this https URL
| Subjects: | Information Retrieval (cs.IR) |
| Cite as: | arXiv:2605.14306 [cs.IR] |
| (or arXiv:2605.14306v3 [cs.IR] for this version) | |
| https://doi.org/10.48550/arXiv.2605.14306 arXiv-issued DOI via DataCite |
Submission history
From: Yuwen Du [view email]
[v1]
Thu, 14 May 2026 03:17:31 UTC (659 KB)
[v2]
Thu, 25 Jun 2026 01:47:21 UTC (2,174 KB)
[v3]
Sat, 27 Jun 2026 03:25:30 UTC (1,695 KB)
来源：HuggingFace Daily Papers（社区热门论文） · arxiv.org