The Research On Chinese Named Entity Recognition Model Based On Cascade Neural Network

Named entity recognition as one of the basic tasks of natural language processing is widely used in the fields of text mining,semantic analysis and machine translation.In today's society where data is becoming more and more massive and heterogeneous,research on the identification of named entities has become one of the key projects in natural language processing at this stage.Most of the existing methods of Chinese named entity recognition rely on prior domain knowledge and have high requirements for researchers' linguistic knowledge.The recognition effect is also not good enough to meet the needs of large-scale engineering applications.With the maturity of deep learning technology in recent years,research on Chinese named entity recognition technology based on artificial neural network has gradually become the main research direction of current Chinese named entity recognition.This paper first analyzes and summarizes the traditional named entity recognition methods and introduces the research results based on artificial neural networks which are identified by foreign named entities.Then based on the traditional named entity recognition method,integrates the word vector and feedforward neural network,and establishes a more flexible named entity recognition framework in this paper.By studying the internal representation of large-scale unlabeled corpus,the system reduces or even neglects the impact of the engineering features,and uses unsupervised methods to identify Chinese named entities.Based on the unsupervised feedforward neural network model,a series of eigenvectors are extracted from multiple target words containing Chinese naming entity features by using PCA dimensionality reduction method.Then these small quantities of prior knowledge are used to realize Semi-supervised named entity recognition,thus further enhancing the recognition effect of the model.The experiments show that the use of feed-forward neural network can greatly improve the problem of recognition relying on prior information and have good recognition speed,but the recognition effect still needs to be improved.Then,after studying the recurrent neural network and the long-short term memory neural network.The paper replaces the original feedforward neural network with the long and short time memory neural network based on the previous feedforward neural network.We also optimize the input vector by optimizing the method of word vector acquisition.The experiments show that the recognition effect of the new model has been further improved,which can be used in unsupervised learning method to obtain more satisfactory results,and to enable them to deal with long sentences compared to the performance of the feedforward neural network to obtain more substantial.Finally,based on the above models,this paper designs a simple prototyping system for Chinese named entity recognition to make it easier to use several models.