Chinese Clinical Named Entity Recognition Based on Stroke-Level and Radical-Level Features

Clinical Named Entity Recognition (CNER) is an important step for mining clini-cal text. Aiming at the problem of insufficient representation of potential Chinese features, we propose the Chinese clinical named entity recognition model based on stroke level and radical level features. The model leverages Bidirectional Long Short-term Memory (BiLSTM) neural network to extract the internal semantic in-formation of Chinese characters (i.e., strokes and radicals). Our method can not only capture the dependence of the internal strokes of Chinese characters, but also enhance the semantic representation of Chinese characters, thereby improving the entity recognition ability of the model. Experimental results show that the accuracy of the model on the CCKS-2017 task 2 benchmark data set reaches 93.66%, and the F1 score reaches 94.70%. Comp ared with the basic BiLSTM-CRF mod-el, the precision of model is increased by 3.38%, the recall is increased by 1.05% and F1 value is increased by 1.91%.