Developing a Medical Question-Answering Chatbot for Egyptian Arabic: A Comparative Study

Abstract

The Egyptian Arabic medical chatbot's development portrays a notable development in the medical field in natural language processing (NLP). This research addresses the Egyptian speech data limitation for question-answering (QA) systems by applying a QA model competent in processing speech and text as input. The model generates answers by implementing the Large Language Model (LLM), creating a dataset for future training models. We implemented Prompt Engineering using the Chain of Thoughts (CoT) approach to improve the model's capability to generate precise Arabic medical answers. Applying the CoT enabled the model to think logically before responding to the question based on the given instructions. We implemented two LLM's: Silma and Phi. The models were evaluated based on BertScore, achieving accuracies of 65.2 % and 59.8 %, respectively. This paper aims to focus on advancing the NLP in the medical Arabic language and contribute a novel dataset for training speech QA models.