MedGapGab at AraHealthQA: Modular LLM Assignment for Gaps and Gabs in Arabic Medical Question Answering
Year: 2025
Type: conference-paper
Abstract: We address Arabic medical question answering (QA) in the AraHealthQA shared task, which evaluates systems on two input formats: (i) fill-in-the-blank terminology items (gaps) and (ii) open-ended patient-doctor dialogues (gabs).We propose MedGapGab, a modular large language model (LLM) framework that assigns each question type to a specialized model-Gemini 2.5 Flash for terminology-focused gaps and DeepSeek V3 for reasoning-intensive gabs.In addition, we use TF-IDF-driven few-shot prompting to retrieve relevant examples from the development set and embed them into the prompts for better contextualization.MedGapGab achieves 87.26% BERTScore, ranking 1 st on the official leaderboard.These results demonstrate that combining TF-IDF-guided example retrieval with type-aware model routing yields strong performance in Arabic medical QA and can inform future work on resource-scarce medical domains.
Author: Baraa Hikal
Institution: University of Göttingen
Language: English
Topic: Topic Modeling
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4416037671