Medical Intention Recognition Based on MCBERT-TextCNN Model

The existing static vector representation methods such as Word2Vec cannot solve the problem that medical text has a long tail concept. In this paper, combining MCBERT medical pre-training model and TextCNN convolutional neural network, a special model for medical text intention recognition, McBert-TextCNN, was proposed. The MCBERT pre-training model uses different mask generation procedures to mask the scope of the token, and introduces two mask strategies: entity mask and whole span mask, which can learn complex terms and various combinations of phrases in the medical field. TextCNN is added to the MCBERT model to extract semantic information features at different levels of abstraction to achieve more accurate text classification effect. The performance of the algorithm is tested on the Chinese medical intention recognition dataset. Experimental results show that the comprehensive evaluation accuracy of the model reaches 0.95, the recall rate reaches 0.94, and the F1 value reaches 0.94, which can effectively improve the effect of medical intention recognition.