Reinforced Hybrid Graph Transformer for Medical Recommendations

Abstract

An enormous amount of heterogeneous Textual Medical Knowledge (TMK), which is crucial to healthcare information systems, has been produced by the explosion of healthcare information. Existing efforts to incorporate and use textual medical knowledge primarily concentrate on setting up simple links and pay less attention to creating computers comprehend information accurately and rapidly. Self-diagnostic symptom checkers and clinical decision support systems have seen a significant rise in demand in recent years. Existing systems rely on knowledge bases that are either automatically generated using straightforward paired statistics or manually constructed through a time-consuming    procedure. The study explored   process to learn textual data, linking disease and symptoms from web-based documents. Medical concepts were scrapped and collected from different web-based sources. The research aims to generate a disease- symptom-diagnosis knowledge graph (DSDKG), with the help of web-based documents. Moreover, the knowledge graph is fed in to Graph neural network with Attention Mechanism (GAT) for learning the nodes and edges relationships. . Lastly Generative Pretrained Transformer 2 (GPT2) all enclosed in a Reinforced learning environment, is used on the trained model to generate text based recommendations.