Research And Application Of Intelligent Recognition Of Instructions Of Traditional Chinese Medicine

Named entity recognition of Chinese medicine instructions refers to the extraction of medical named entities that are closely related to Chinese medicine from the text of Chinese medicine instructions.However,the long text of traditional Chinese medicine instructions,strong professional vocabulary,entity aggregation and nesting,and imperfect domain dictionaries make the named entity recognition method based on deep learning unable to capture long-distance dependencies,insufficient semantic extraction capabilities,and inability to use lexical information.In this regard,this paper studies two methods of named entity recognition based on word-level input and named entity recognition based on word information fusion,and makes the following work and contributions.Part of the public data sets and the text of the Chinese medicine instructions published on the medical website are used as data sources,and the entity types and labeling rules that need to be identified in the Chinese medicine instructions are agreed,and the original data is re-labeled by using deep learning and artificial combination.Under the guidance of relevant experts,the preliminary marked corpus was reviewed,and the construction of the corpus of Chinese medicine manuals was completed.Aiming at the problems that the BiLSTM model cannot capture the long-distance dependence,the semantic extraction ability is insufficient,and the recognition effect is not good in the named entity recognition task of the Chinese medicine instructions,a named entity recognition method for the Chinese medicine instructions based on the deep self-attention BiLSTM model is proposed.First,add a multi-layer mapping module to the BiLSTM model to increase the depth of the model to extract deep semantic features,then add a dynamic activation function to enhance the model’s fitting ability,and finally add a self-attention mechanism to capture long-distance dependencies,which improves the BiLSTM model in The recognition effect in the task.In view of the imperfect domain dictionary and the inability of the deep selfattention BiLSTM model to utilize lexical information.Furthermore,a word information fusion method is proposed.This method does not require a domain dictionary,extracts semantic information from continuous word vectors as lexical features,and uses feature distillation and Reshape mechanisms to retain effective lexical features and make full use of them to further improve the performance of the deep selfattention BiLSTM model in tasks.Finally,the experimental verification is carried out on the corpus of Chinese medicine manuals.The experimental results show that the F1 value of the BiLSTM model based on deep self-attention is increased by 4.18% compared with the BiLSTM model.After combining the word information fusion method with the BiLSTM model based on deep self-attention,the F1 value is increased by 4.98% to 83.02%,which proves that The effectiveness of the proposed method realizes the accurate identification of named entities in the instructions of traditional Chinese medicine,and provides a certain guarantee for the downstream tasks in the field of medicine.