DRUG INFORMATION EXTRACTION FROM PRESCRIPTION WITH GCN

Abstract

In this thesis, We delve into a comprehensive study on the application of Graph Convolutional Networks (GCN) for prescription extraction, specifically focusing on extracting drug names to meet the needs of health monitoring and drug information retrieval. Here, We adopt an approach that harnesses the graph structure within prescription data to capture relationships between information positions within the prescription, specifically the relationships between crucial entities that need to be extracted, such as drug names. The core idea of the project is closely tied to the design, training, and development of the GCN model on the labeled prescription dataset. The model is specifically designed to learn and generalize information related to drug names, enabling the extraction of drug name information. Additionally, the project aims for generality in the prediction process to enhance accuracy in predictions. The outcomes of the project hold significant implications for healthcare management and health care, thanks to the progress in applying machine learning in real-life situations. In summary, this research not only broadens the perspective of applying machine learning to natural language processing and information extraction but also brings significance to the field of health informatics, laying the groundwork for future breakthroughs and further improvements in both academic learning and real-world applications. Experimental results revealed that our model has achieved 0.98%, 0.95%, 0.89% and 0.92% for accuracy, precision, recall and F1-score. With this accuracy, the project has met the goal of extracting drug name information from invoices and is a premise for other projects to extract specific information based on images.