An Autonomous Dual-channel Entity Recognition Method for Chinese Acupuncture

Due to the strong specialization of knowledge in the field of acupuncture and moxibustion, the entities are more closely related, the structure is complex, and there are more mixed words in Chinese and English, which leads to the poor effect of the general entity recognition model in the field of acupuncture and moxibustion. In this regard, a two-way meridian entity recognition model is proposed: for the entity categories of acupoints, manipulation techniques, and evidence types that are more fixed and contain corresponding keywords, we adopt the incorporation of lexicon information into the bottom layer of BERT to enhance the effective utilization of lexicon information, and we introduce a contribution factor in BiLSTM to improve the correlation before and after the statements and enhance the semantic information; for the diseases and symptoms that have large vocabularies and are characterized by a large amount of abbreviations, an extra attention mechanism is introduced in BERT to obtain word-level features and semantic information in different dimensions, to improve the shortcomings of the deep semantic information output from BERT which lacks the underlying word-level features; finally, an integration of the results of the two-way warping is done. In order to verify the effectiveness of the proposed model, it is compared with the existing methods on the labeled dataset, and the F1 value of the model on the dataset can reach 94.21%. The experimental results prove that the proposed dual-road meridian entity recognition model can recognize entities in the field of acupuncture and moxibustion better than other methods.