NCU-IISR: Enhancing Biomedical Question Answering with GPT-4 and Retrieval Augmented Generation in BioASQ 12b Phase B

In this paper, we introduce our system and submissions in BioASQ 12b phase b [1], highlighting a significant improvement with GPT-4 and the integration of Retrieval Augmented Generation (RAG) techniques. We describe our prompt engineering methods and the experimental procedures followed. Because GPT-4 has proven effectiveness in generating answers and its ability in the biological domain, our system utilizes GPT-4 to address biomedical question-answering (QA). Leveraging OpenAI's ChatCompletions API, we refined previous prompt engineering approaches [2] for BioASQ 11b phase b. This year, the addition of RAG techniques significantly improved the information retrieval capabilities of our system. Consequently, our latest submission employed what we experimented to be the most effective prompts and techniques, achieving excellent performance across multiple metrics in the fourth batch.