文章背景与核心概要
随着学术研究投稿量的迅猛增长，传统的人工审稿能力已显得捉襟见肘，利用大语言模型（LLM）辅助进行论文与代码之间的差异检测应运而生。然而，现有的单智能体LLM范式普遍面临上下文容量受限和检测视角单一的问题，导致召回率较低。
为了克服这些局限，研究人员推出了 Dude——首个专为识别学术论文与其对应代码实现之间差异而设计的双重检测多智能体系统。作者发现，论文语言与代码语言之间固有的粒度不对称性，常常会在多智能体框架中引发过度解释和过度汇报，从而导致较高的误报率。为了解决这一痛点，Dude 创新性地引入了：1. 粒度对齐协商机制；2. 两阶段显著性过滤机制。在真实世界的论文-代码差异数据集上的实验评估表明，与基线方法相比，Dude 的召回率和准确率显著提升高达 22.8%，F1 分数提升高达 18.7%。
Dude: A Dual-Detection Multi-Agent System for Paper-Code Discrepancy Detection
Summary
As the volume of research submissions rapidly outpaces human review capabilities, Large Language Model (LLM)-empowered systems have emerged to assist with paper-code discrepancy detection. However, existing single-agent LLM paradigms suffer from limited context capacities and one-sided detection, resulting in low recall.
To overcome this, researchers introduce Dude, the first Dual-Detection Multi-Agent System designed specifically to identify discrepancies between academic papers and their corresponding code implementations. The authors discovered that the inherent granularity asymmetry between paper-language and code-language often leads to over-interpretation and over-reporting in multi-agent frameworks, causing high false-positive rates. To combat this, Dude introduces: 1. Granularity-Aligned Negotiation, and 2. A Two-Stage Salience-Filtering Mechanism.
Experimental evaluations on real-world paper-code discrepancy datasets demonstrate that Dude significantly enhances both recall and precision by up to 22.8% and improves the F1 score by up to 18.7% compared to baseline methods.
Paper Metadata / 论文元数据
- arXiv Identifier: arXiv:2609.03416 [cs.AI]
- Primary Subject: Artificial Intelligence (
cs.AI)- Secondary Subject: Machine Learning (
cs.LG)- Submission Date: September 3, 2026
- Conference Acceptance: Accepted to EMNLP 2026 Main Conference
- Authors:
- Weijie Liu
- Running Zhao
- Wenhao Yuan
- Jinfeng Xu
- Zhanfeng Xu
- Xiaoxi Zhang
- Edith Cheuk-Han Ngai
Abstract / 摘要
LLM-empowered paper-code discrepancy detection has received growing concern since the scaling of research submissions exceeds the manual review capability. However, the limited context capacity and one-sided discrepancy detection of existing single-agent LLM paradigms lead to an inferior recall performance in detecting discrepancies. In this paper, we propose Dude, the first Dual-Detection Multi-Agent System for paper-code discrepancy detection. We discover that the granularity asymmetry of the paper-language and code-language introduces over-interpretation and over-reporting challenges in a multi-agent system design for discrepancy detection, resulting in increasing false positives. To address this, we propose a granularity-aligned negotiation and a two-stage salience-filtering mechanism in Dude, which effectively prevents agents from falsely reporting discrepancies. Experimental results in real-world paper-code discrepancy datasets showcase Dude's significant recall and precision improvement by up to 22.8%, increasing F1 score by up to 18.7% compared to baseline methods.