Year: 2025
Type: conference-paper
Abstract: This paper presents our participation in the QIAS 2025 shared tasks, namely Islamic Inheritance Reasoning and Islamic Knowledge Assessment sub-tasks.We propose an Islamic Retrieval-Augmented Generation (RAG) system that integrates multiple knowledge sources and semantic retrieval methods.Our evaluation compares multilingual general-purpose models and Arabiccentric models, using the accuracy metric.Results show that multilingual models consistently outperform Arabic-language models.The Mistral-large achieved the highest accuracy in Task 1 (72%) using basic RAG with our augmented knowledge resource, while GPT-4o with RAG and K2R retrieval achieved the best score in Task 2 (87.71%).These findings highlight the effectiveness of RAG in enhancing LLM performance for complex Islamic reasoning and knowledge assessment tasks.
Author: Sanaa Alowaidi
Institutions: University of Leeds, King Abdulaziz University
Language: English
Topic: Topic Modeling
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4416037613