Research On Named Entity Recognition For Diagnosis And Prevention Of Aquatic Animal Diseases

Aquatic animal diseases are the key to the rapid development of aquaculture industry.Effective disease diagnosis and prevention requires accurate representation,and the knowledge of aquatic medicine is fully utilized.As a widely used knowledge representation and application tool,knowledge graph can effectively solve the aquatic medical knowledge representation,and then provide a new means for the diagnosis and prevention of aquatic animal diseases.Named entity recognition is the foundation of constructing aquatic medicine knowledge graph.The accuracy of named entity recognition directly affects the quality of graph construction.However,the recognition accuracy of named entities is affected by the large number of nested entities in aquatic medicine named entities,resulting in low recognition accuracy.The construction quality of aquatic medicine knowledge graph is also affected.This paper aims at the above problems,chinese named entity recognition for the diagnosis and prevention of aquatic animal diseases is carried out.The main research contents and innovations are as follows:(1)Constructing a corpus of aquatic medicine.In order to reduce the impact of data on the training of aquatic medicine entity recognition model,an aquatic medicine corpus for aquatic medicine named entity recognition was constructed by cooperating with experts in the field of aquatic medicine and referring to the construction method of datasets in the general field.The corpus covers a comprehensive aquatic medicine knowledge can effectively improve the training effect of the model and improve the generalization ability of the model.(2)An entity recognition method based on multi-kernel convolution neural network is proposed.In order to reduce the influence of the position vector on the effect of entity recognition,the BERT model is introduced to add entity location information;in addition,in view of the problem that the entity features are not obvious during the entity recognition process,multi-kernel convolution is used to extract the entity features step by step,and the extracted feature matrix is combined with.The input feature matrix is fused to enhance the feature representation of the entity;the discriminant layer is set in the output process to discriminate and process the prediction results,and realize the output of multiple prediction results of the entity.Experiments on the aquatic medicine corpus show that the proposed method can effectively improve the recognition effect of aquatic medical named entities,and the F1 value reaches 88.48%,which is 2.74% higher than the BERT+Bi LSTM+ATT+CRF model with better recognition effect in the field.percent.(3)This paper proposes a named entity recognition method for aquatic medicine combining BERT and Ca Bi LSTM.The Ca Bi LSTM model is designed by adopting the "layered idea" for nested entity recognition,and the features of the inner layer entities are used to improve the discrimination of the outer layer entities,so as to realize the accurate recognition of the outer layer entities,and solve the problem of low recognition accuracy caused by nested entities.Introduce the BERT model to add entity location information,so that the same entity has different features in different locations,and solve the problem of polysemy resulting in low recognition accuracy;the test results show that the proposed method can effectively identify the corpus of aquatic medicine.Compared with the proposed multi-kernel convolution neural network-based named entity recognition method,the F1 value reaches 92.96%,an increase of 4.48 percentage points.