Master's Thesis Volkan Tatlikazan

Motivation

Medical QA systems of the day lack the choice for users to select their source of medical information in any category (e.g. Research, Education, Clinical, Consumer). As medical information is accepted after many long winded theoretical/practical research and years of application in real life, it is highly cultural and depends on local multi-generational information. By giving the user the freedom to select their information source and making the LLMs and additional RAG methods easy to understand and use, we hope to achieve increased understanding of the reasoning and trust on the medical processes.

Goal

The goal of this research is to develop systems that provides evidence depended in medical situations. For this we go after the following research questions:

• RQ1: What is the best performing approach for medical question answering and do these approaches generalize well over diverse (or unseen) datasets for medical question answering?
• RQ2: How can we generate answers to medical questions using retrieved medical evidence (or knowledge) using LLMs and methods like RAG (Retrieval-augmented generation)? 
• RQ3: Can we generate medically accurate explanations in a Q&A format for users to understand medical information easier?