Athar at QIAS2025: LLM-based Question Answering Systems for Islamic Inheritance and Classical Islamic Knowledge

Résumé

The intersection of Arabic linguistic complexity and specialized reasoning presents a key challenge for Islamic question-answering systems, particularly in the under-addressed area of inheritance law.This paper presents our methodology for the QIAS2025 shared task, assessing LLM capabilities in Islamic knowledge through two subtasks: Inheritance Reasoning (ʿilm al-mawārīth) and General Islamic Assessment.A zero-shot, prompt-based approach with DeepSeek-R1 (deepseek-reasoner) addresses the former, while a three-stage RAG pipeline handles the latter.Our approaches achieved competitive results, with an accuracy of 0.704 for inheritance reasoning (10th place/15 teams) and 0.9272 for general Islamic assessment (2nd place/10 teams), demonstrating the efficacy of tailored model strategies for religious QA.These insights pave the way for more culturally and linguistically adaptive AI systems in Islamic scholarly applications.

Citer ce document

Accès au document

Texte intégral en lecture en ligne, réservé aux abonnés SPHAERO et aux membres de l'institution. Se connecter

Auteur(s)

Statistiques

Consultations : 1

Téléchargements : 0