Research On Chinese Named Entity Recognition Based On Local Adversarial Transfer Training

The purpose of named entity recognition is to identify the names of people,place names,organization names and other proper nouns or meaningful specific entities in the text,and it is the basis of knowledge map,text generation,text extraction and other downstream tasks.In order to improve the generalization ability of Chinese named entity recognition model,it usually needs to spend a lot of manpower and material resources to label massive data.Using transfer learning technology to transfer information between different domains can help the model learn the language rules of the target domain and reduce the model’s dependence on annotated data.The current Chinese named entity recognition model has a good performance in plane entity recognition,but due to the complex semantic composition of Chinese words,entity nesting,word boundary ambiguity and other problems,there are still technical problems in Chinese nested named entity recognition.Under the task of Chinese named entity recognition,this essay combines transfer learning with adversarial training,makes use of the learned language rules to supplement the features of the current task,and fine-adjusts the pre-training language model to improve the model stability.On this basis,machine reading comprehension technology is added to improve the use of context information,and further improve the effect of Chinese nested named entity recognition.The research content mainly includes the following two points:1.Named Entity recognition based on local adversarial training and transfer learningIn order to solve the problem that a large amount of annotation data is needed and the model generalization ability is not strong in named entity recognition,a named entity recognition model combining transfer learning and local adversarial training is proposed in this essay.Based on the ALBERT pre-training language model,perturbation factors are added to the ALBERT embedding layer to form local adduction training.The generated adduction samples can improve the robustness of the model.Then,the output matrix obtained is extracted by Bi LSTM features and combined with the attention mechanism for feature weighting and label constraints.By comparing the experimental results,the accuracy of the proposed model is better than that of the comparison model,which proves that the proposed method can effectively improve the performance of named entity recognition model.2.Research on Chinese nested named Entity Recognition based on Machine Reading comprehensionIn order to solve the problem of Chinese nested named entity recognition,a named entity recognition model combining machine reading comprehension and local adversarial transfer training is proposed in this paper.By using machine reading comprehension technology,entity recognition task is transformed into knowledge question answering task,triplet is constructed to solve the problem of mutual nesting among entities,and then the triplet information is obtained by ALBERT pre-training layer to get the output matrix,and the features extracted by Bi LSTM-Attention filter.Finally,the binary classification mechanism is used to judge the start and end position labels of the answer and predict the physical answer.Through comparison of experimental results,the accuracy of this model is improved compared with the model without machine reading comprehension method,which proves that machine reading comprehension technology is helpful to solve the problem of Chinese nested entity.Figure [20] Table [13] Reference [69]...