Quasar at AraHealthQA Track 1 : Leveraging Zero-Shot Large Language Models for Question and Answer Categorization in Arabic Mental Health
Year: 2025
Type: conference-paper
Abstract: Pre-trained language models (PLMs) show potential for advancing mental health care, yet their effectiveness in Arabic mental health contexts is underexplored.This study evaluates PLMs on two multi-label classification tasks from the AraHealthQA 2025 shared task Track 1: question categorization and answer strategy classification.We systematically evaluate several LLMs spanning Arabic-specialized, multilingual, and general-purpose architectures using zero-shot inference, with comparative analysis revealing Qwen3-14B's superior performance.Our approach combines promptbased inference, label mapping, and strategically crafted Arabic prompts.Experiments on 350 training and 150 test samples demonstrate competitive performance, securing 4 th place in both tasks (Question F1: 0.52, Answer F1: 0.76; Question Jaccard: 0.41, Answer Jaccard: 0.66).These findings reveal strengths and limitations of current PLMs for detecting complex intents in Arabic mental health contexts.
Authors: Adiba Fairooz Chowdhury, Md Sagor Chowdhury
Institution: Chittagong University of Engineering & Technology
Language: English
Topic: Mental Health via Writing
Subfield: Social Psychology
Field: Psychology
Domain: Social Sciences
Open Access status: hybrid
OpenAlex ID: W4416037815