Research On Named Entity Recognition Based On Neural Network Ensemble

With the development and popularity of e-commerce and social networks,natural language text information that needs to be processed and analyzed presents a trend of geometric progression.Named entity recognition is a solid foundation and an important component of natural language applications such as information extraction,public opinion analysis,question and answer system,information retrieval,syntax analysis,machine translation,etc.,and has received much attention as a natural language processing task.In recent years,with the success of neural networks in the fields of image recognition and speech recognition,neural networks have effective application in named entity recognition,either.Usually,the word vector space is usually used to mapping the word representation as a fixed global feature.Each vocabulary uses a specific vector representation,and the training and test results are performed through a single neural network.The feature extraction is single,and the generalization ability is poor.In order to improve the performance of named entity recognition feature extraction and solve the problem that different meanings of polysemous words in word representation are represented by the same vector,this thesis proposes an end-to-end multi-prototype representation,without subjective manual intervention or beforehand selection.The multi-prototype representations are extracted while training.The feature extraction of words is closer to natural language characteristics,and the performance of feature extraction in named entity recognition is improved.In order to solve the problem that the single neural network model has limitations in named entity recognition,a neural network ensemble model with diversity is proposed.The difference between the base classifiers is measured by measuring the correlation of the error distribution of the base classifiers with statistical indicators.Then,the indicators are used to measure the benefit of ensemble learning the model disturbance brings.After studying the generation of diversity of the neural network model in the named entity recognition,the differential base classifier is trained unreasonably with model disturbance to guarantee diversity of ensemble learning and to improve the performance of the neural network ensemble.In addition,the insensitive loss function is introduced into the bidirectional long short-term memory neural network,and it is proved that it can guarantee the difference between individual classifiers.