Arabic Large Language Models for Medical Text Generation

Summary

This study explores fine-tuning large language models (LLMs) for Arabic medical text generation to improve hospital management systems. A unique dataset was collected from social media, capturing medical conversations between patients and doctors, and used to fine-tune models like Mistral-7B, LLaMA-2-7B, and GPT-2. The fine-tuned Mistral-7B model outperformed the others with a BERT F1-score of 68.5%. Why it matters: The research demonstrates the potential of generative AI to provide scalable and culturally relevant solutions for healthcare challenges in Arabic-speaking regions.