Medical Text Entity Recognition Based on Deep Learning

Abstract

Named entity recognition is a very important basic task in natural language processing, and a basic technology for many high-level applications of natural language processing. Traditional methods to solve named entity recognition are mainly based on rules and statistics. The rule-based method requires strong linguistic knowledge and is poor in generality. Based on the above background, the purpose of this paper is to recognize medical text entities based on deep learning. This paper proposes a new LSTM framework that integrates dual-channel and sentence-level reading control gates. In the input part, double channels are added to obtain two kinds of semantic information from static word vectors and fine-tuned word vectors. Then, the read-in control gate is integrated inside the neural network to determine the propagation of the sentence representation vector. Finally, the CRF model is used to fully consider the dependency relationship between the types of context words when outputting tags. The F1 value of 89.49% was achieved on the BioCreative II GM corpus. In summary, this article applies two deep learning methods to improve the performance of biomedical named entity recognition tasks. Finally, without adding any artificial features and post-processing operations, this article has achieved 89.94% F1 value on the BioCreative II GM corpus, and it is 0.89% higher than the F1 value of the current best system.