Hamyaria at PalmX2025: Leveraging Large Language Models to Improve Arabic Multiple-Choice Questions in Cultural and Islamic Domains
Year: 2025
Type: conference-paper
Abstract: Large language models (LLMs) have been widely used recently.Adapting these models to multiple languages would enhance the accuracy and precision of the other languages.Applying LLMs with Arabic language could improve the prediction of Arabic language.This work applies LLMs with MCQs of Arabic in both the cultural and Islamic domain.The dataset used is PalmX, which is an MCQ benchmark dataset.In this work, traditional and AI generation data augmentations are used.For the cultural domain, we applied data augmentation techniques, including paraphrasing using Fanar-1-9B-Instruct model and answer shuffling.For the Islamic domain, we used the original dataset without augmentation to maintain content integrity.We then fine-tuned the Qwen2.5-3B-Instructmodel on both datasets and evaluate its performance, achieving 65.90% accuracy on the cultural set and 70.83% on the Islamic set.Experiment and evaluation are discussed and the best accuracy achieved in this work is explained in both domains.
Authors: Walid Al-Dhabyani, Hamzah A. Alsayadi
Institutions: Cairo University, Hadhramout University, Ibb University
Language: English
Topic: Topic Modeling
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4416037608