Research On Named Entity Recognition Of Court Judgment Documents Based On Deep Learning

With the continuous advancement of judicial reform in China,the judiciary is facing a new situation in which the trial business is complicated and the cases are accumulating.By studying the identification of named entities in the judgment documents,the user's understanding of the content of the irregular text and the grasp of the key content can be accelerated.It lays the foundation for the subsequent evidence identification,plot determination,judgment result analysis and program compliance.The current deep learning technology has been widely used in natural language processing and has achieved good results.In this paper,deep learning is used in the study of court judgment documents.Based on the long-short memory neural network(LSTM)based on deep learning,the LSTM model with the conditional random field(CRF)transfer layer is proposed.Training gets a model with contextual information that is more in line with the linguistic features of textual information.Since there is no open normative corpus in the field of refereeing documents in the world,this paper learns the LSTM model through a large number of standard corpus in the news field.On the one hand,the linguistic knowledge of the news field is loaded into the referee through the method of deep learning to load the pre-training word vector.The corpus of the instrument is superimposed on the training set for iterative training,and the model of the decision instrument field is selected.The F value of the final model reaches 0.807.The innovation of this paper is to use the idea of migration learning to identify the cross-domain named entities and achieve better results.