Research On Named Entity Recognition Method For Traditional Chinese Medicine

The digital development of traditional Chinese medicine has been accelerated by the extensive application of artificial intelligence technology and the strong support of national policies.The digital development of traditional Chinese medicine is not only a need for people’s livelihood,but also a move to promote the excellent national culture.Tasks such as the construction of the Knowledge Graph of traditional Chinese medicine and the construction of the dialogue system for seeking medical consultation are hot research directions.Whether the above tasks can be carried out smoothly and efficiently,the Named Entity Recognition technology plays a pivotal role.Named Entity Recognition technology aims to identify key entities from traditional Chinese medicine,and faces many challenges,which is the cornerstone of the digital development of traditional Chinese medicine.In addition to the inherent weaknesses of Chinese named entity recognition itself,Chinese medicine texts are highly specialized,and the texts are more obscure and difficult to understand than vernacular texts.In response to these pain points,this thesis has done the following work:1.Build a standard dataset.Use crawler and other technologies to search for raw data such as records and manuals of traditional Chinese medicine.After data cleaning,manual and automatic labeling is used.There are eight kinds of entity,including drug names,drug ingredients,drug properties,drug flavors,drug dosage forms,drug efficacy,diseases,symptoms,and syndrome.2.Propose two named entity recognition methods based on pre-trained language models.Entities in the field of traditional Chinese medicine are all professional terms,and it is difficult and critical to capture the semantic information.Pre-trained language models help to enhance semantic representation,resulting in more accurate text feature extraction.3.Optimization for nested entities.The existence of nested entities greatly affects the recognition efficiency.For this reason,this thesis uses the affine transformation technology which commonly used in the field of computer vision,and improves it to reduce the interference of nested entities on named entity recognition from the level of rich semantic representation.Finally,a full experiment is carried out on the self-built dataset.The results show that before the optimization of nested entities,the recognition effect of drug efficacy entities is the best,and its precision rate,recall rate and F1 value reach 93.11%,92.45% and 92.78%respectively.After the fusion of the affine transformation mechanism,the recognition effect of various entities is enhanced,and the accuracy of the drug ingredients entity is increased by 8.23%.After analyzing the experimental data,it is concluded that the named entity recognition method proposed in this thesis for the field of traditional Chinese medicine is feasible and effective.