A Novel Approach for Persian Medical and Drug Question-Answering using Large Language Models

Anonymous

Abstract:

The remarkable progress of large language models (LLMs) in comprehending and responding to human instructions has captivated attention in recent years. However, their performance is predominantly optimized for English, posing limitations when applied to the medical domain. Consequently, their precision in critical tasks such as diagnoses, drug recommendations, and medical advice may fall short. Moreover, the complexities associated with training and deploying dialogue models in hospital settings have hindered the widespread adoption of LLMs.
To address these challenges, this paper presents a comprehensive approach that involves data collection, model training, and comparative analysis.
We gather a specialized dataset in Drug-based datasets.
To accommodate a broader user base, we facilitate question input in both Persian and English languages. For Persian inquiries, we leverage our dataset and employ the ChatGPT model to translate them into English, ensuring uniformity in language processing.
Subsequently, we employ various techniques to identify the most suitable answer. This includes leveraging different pre-trained language models. The objective is to generate accurate and contextually relevant responses to user queries.
Upon obtaining the answer in English, we translate it back to Persian, ensuring effective communication with Persian-speaking users.
Through our comprehensive approach encompassing data collection, model training, translation, and answer generation, we aim to overcome the limitations faced by LLMs in the medical domain. By optimizing their performance for medical tasks and facilitating multilingual communication, we strive to enhance precision and effectiveness in diagnoses, drug recommendations, and medical advice. This research paves the way for the wider adoption of LLMs in healthcare, ultimately benefiting both patients and medical professionals.