The Study of Named Entity Identification in Chinese Electronic Medical Records Based on Multi-tasking

A Bidirectional Long Short-Term Memory Conditional Random Feld based on a combination of word segmentation task and named entity recognition task is proposed to address the problem of named entity recognition in structured electronic medical records. This model enriches the feature set of named entity recognition tasks by incorporating shared LSTM to capture word boundary information in word segmentation tasks, thereby achieving the effect of improving named entities. The experimental data collection consists of a discharge summary of 500 coronary heart disease patients and 2000 cardiovascular disease patients provided by a tertiary hospital in Guangdong Province. Comparedwith other models in electronic medical record entity recognition tasks, the multi-task learning model based on Bi-LSTM-CRF achieved an F-measure value of 0.927. The experiment shows that the multi-task electronic medical record entity recognition model based on Bi-LSTM-CRF can effectively learn information from multiple related tasks, which well meets the practical needs of clinical practice.