Binary_Bunch at AraHealthQA Track 1: Arabic Mental Health Q&A Classification Using Data Augmentation and Transformer Models
Year: 2025
Type: conference-paper
Abstract: Mental health question-answering (Men-talQA) is essential for delivering accessible and reliable mental health support.Natural language processing (NLP) techniques are increasingly integral to such systems, enabling automated categorization of questions and answers to improve information retrieval, response accuracy, and user guidance.In AraHealthQA 2025 (Track 1), we addressed two subtasks: multi-label question categorization and answer categorization.We proposed an XLMR-Arabic pipeline enhanced with a two-stage data augmentation strategy, combining large language model (LLM)-based paraphrasing with synthetic label merging.Additionally, we evaluated the effectiveness of fine-tuned multilingual transformers, LLMs adapted with low-rank adaptation (LoRA), and LLMs under few-shot settings.Experimental results show that XLMR-Arabic achieved the best performance, reaching Jaccard scores of 53% and 77.44% on Subtasks 1 and 2, respectively, ranking our team second in both tracks.
Authors: Sajib Bhattacharjee, Ratnajit Dhar, Kawsar Ahmed, Mohammed Moshiul Hoque
Institution: Chittagong University of Engineering & Technology
Language: English
Topic: Topic Modeling
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4416037773