arXiv 2601.075282026-06-23cs.CLcs.AI

从RAG到智能体RAG：面向可靠的伊斯兰问答

From RAG to Agentic RAG for Faithful Islamic Question Answering

Qatar Computing Research Institute, HBKU, Qatar（卡塔尔计算研究中心，HBKU，卡塔尔）；College of Humanities and Social Sciences, HBKU, Qatar（人文与社会科学学院，HBKU，卡塔尔）；Arab Center for Research and Policy Studies, Qatar（阿拉伯研究中心与政策研究所，卡塔尔）；College of Islamic Studies, HBKU, Qatar（伊斯兰研究学院，HBKU，卡塔尔）；College of Public Policy, HBKU, Qatar（公共政策学院，HBKU，卡塔尔）

Gagan Bhatia, Hamdy Mubarak, Mustafa Jarrar, George Mikros, Fadi Zaraket, Mahmoud Alhirthani, Mutaz Al-Khatib, Logan Cochrane, Kareem Darwish, Rashid Yahiaoui, Firoj Alam

备注：: Islamic Question Answering; Faithful Question Answering; Retrieval-Augmented Generation; Agentic RAG; Large Language Models

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

针对LLM在伊斯兰问答中的幻觉与弃权问题，构建双语基准IslamicFaithQA，并提出基于结构化工具调用的智能体RAG框架，显著提升正确性与鲁棒性。

摘要：

大型语言模型（LLM）越来越多地被用于伊斯兰问答，其中无根据的回答可能带来严重的宗教后果。然而，标准的MCQ/MRC式评估（MCQ：多项选择题，MRC：机器阅读理解）未能捕捉关键的真实世界失败模式，特别是自由形式的幻觉和在证据不足时放弃回答的能力。为弥补这一差距，我们引入了IslamicFaithQA，一个包含3,810个项目的双语（阿拉伯语/英语）生成式基准，具有原子单金答案，能够直接测量幻觉和弃权。此外，我们开发了一个端到端的基于文本的伊斯兰建模套件，包括（i）25K阿拉伯语文本基础的SFT推理对，（ii）5K双语偏好样本用于奖励引导对齐，以及（iii）一个包含约6k个原子节（ayat）的节级《古兰经》检索语料库。基于这些资源，我们开发了一个智能体古兰经基础框架（智能体RAG），该框架使用结构化工具调用进行迭代的证据寻求和答案修订。在阿拉伯语中心和多语言LLM上的实验表明，检索提高了正确性，并且智能体RAG在标准RAG之上取得了最大的增益，即使使用小型模型（即Qwen3 4B）也能实现最先进的性能和更强的阿拉伯语-英语鲁棒性。我们已将数据集公开。此 https URL

英文摘要：

Large Language Models (LLMs) are increasingly used for Islamic question answering, where ungrounded responses may carry serious religious consequences. Yet standard MCQ/MRC-style evaluations (MCQ: Multiple choice questions, MRC: Machine Reading Comprehension) do not capture key real-world failure modes, notably free-form hallucinations and the ability to abstain when evidence is insufficient. To address this gap, we introduce IslamicFaithQA, a 3,810-item bilingual (Arabic/English) generative benchmark with atomic single-gold answers, which enables direct measurement of hallucination and abstention. We additionally developed an end-to-end grounded Islamic modeling suite consisting of (i) 25K Arabic text-grounded SFT reasoning pairs, (ii) 5K bilingual preference samples for reward-guided alignment, and (iii) a verse-level Qur'an retrieval corpus of ~6k atomic verses (ayat). Building on these resources, we develop an agentic Quran-grounding framework (agentic RAG) that uses structured tool calls for iterative evidence seeking and answer revision. Experiments across Arabic-centric and multilingual LLMs show that retrieval improves correctness and that agentic RAG yields the largest gains beyond standard RAG, achieving state-of-the-art performance and stronger Arabic-English robustness even with a small model (i.e., Qwen3 4B). We made the datasets are publicly available. https://huggingface.co/datasets/QCRI/IslamicFaithQA

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。