TCE at IslamicEval 2025: Retrieval-Augmented LLMs for Quranic and Hadith Content Identification and Verification
Published in ArabicNLP2025 (EMNLP2025), IslamicEval 2025, 2025
Given the inherent nature of LLMs to hallucinate factual content, the research community developed methodologies for factual verification. However, LLMs are not just limited to hallucinating facts; Arabic religious texts, namely, the Holy Quran and Hadith, were found to be altered by contemporary LLMs, violating their meaning and spreading misconceptions. In this research, we propose a framework to detect and cross-check religious content in generated Arabic text. Our system achieves 86.11% accuracy for segment detection and 89.82% accuracy in content verification from authoritative sources.
Recommended citation:
Download Paper