NLP and GPT Integration for Drug Information Analysis and Health Advice

As China continues to develop, people are now paying more attention to getting good health care. It is, therefore, fundamental to get proper and qualified medical advice and treatment. However, there are times when doctors or pharmacists are not readily available to dispense advice on medications taken by the patient. In this paper, we explore the possibility of integrating Generative Pre-trained Transformer (GPT) models and Named Entity Recognition (NER) methods. Existing NER methods face greater challenges when applied on Chinese drug specification due to differences in syntax and semantics. We used Model Optimization based on Adversarial Training, Projected Gradient Descent, Model Optimization based on Mixed Precision Training, and Model Optimization based on Semi-supervised Learning to improve the accuracy of identifying keywords in the drug specifications. These keywords were then fed into different GPT models to generate relevant health advice. Although the results show only marginal improvement in keywords identification using some of the NER methods, the integration to GPT models is promising and deserves further investigation.