Application of Named Entity Recognition by Self-Attention BiLSTM-CRF to Chinese Prescription Document

Abstract

Reporting the contents of the prescription is a time-consuming and labor-intensive task for pharmacists, and it is easy to make mistakes in this kind of work that requires repeated confirmation under the time pressure of customers. Because medicines usually have an impact on the human body, if the wrong medicine is accidentally provided to the customer, it may even lead to death. Therefore, in our work, we train a bidirectional LSTM-CRF network combined with an attention mechanism to perform the NER task on prescription notes and extract key prescription declaration information. The semi-automatic labeling module was implemented using a rule-based approach, and a total of 636 different prescription notes were trained and tested, including hospitals, clinics, and medical centers. The identification accuracy rate is more than 90 % in each field of declaration information. Provides a UI interface for subsequent identification to facilitate pharmacists to modify the identification content and declare key information.