Research on Chinese Named Entity Recognition Based on Neural Network

Abstract

Named entity recognition is a fundamental and important task in the field of natural language processing. The effect of existing solutions in practical applications do not meet need of people. Firstly, this paper studies the scheme of recurrent neural networks with conditional random fields. Further, a series of experiments were performed on the input word vector dimension, the number of neural nodes within the neural network, and the variant structure of neural networks. Finally, a method of combining neural network with attention mechanism is proposed. An attention layer is built over the output of the recurrent neural network and the F1 score reaches 91.28%.