MindLLM at AraHealthQA 2025 Track 1: Leveraging Large Language Models for Mental Health Question Answering
Year: 2025
Type: conference-paper
Abstract: This paper presents our submission to the Ara-HealthQA 2025 shared task (Alhuzali et al., 2025), Sub-task 3: Arabic Mental Health Question Answering.We evaluated four large language models-GPT-4o, Gemini, Allam, and Qwen-using various prompting strategies.A simple 3-shot prompt, instructing the model to respond in Arabic, consistently outperformed zero-shot, 5-shot, and more complex methods.GPT-4o achieved the best results, with a BERTScore F1 of 0.670 on the official hidden test set, ranking 2nd overall.The system required no fine-tuning or external data, relying solely on prompt design and consistent evaluation.
Author: Nejood Abdulaziz Bin Eshaq
Language: English
Topic: Topic Modeling
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4416037799