Research on Text Summarization Generator Method Based on Input Text Linguistic Features and Copy Mechanism

Abstract

The Sequence-to-Sequence (Seq2Seq) of neural network (NN) method based on recurrent neural network (RNN) and attention mechanism plays an important role in information extraction and automatic summarization. However, this method can’t make full use of the text linguistic feature information, and there is a problem of unregistered words in the generated results, which affects the accuracy and readability of the text summarization. To solve this problem, the text linguistic feature is used to improve the input characteristics, and the copy mechanism is introduced to alleviate the problem of unregistered words in the summarization generation process. On this basis, a new method named Copy-Generator Model (CGM) based on Seq2Seq model is proposed to improve the effect of text summarization. The results of experiments using LCSTS (Large Scale Chinese Short Text Summarization) as the data source show that the proposed method in this paper can effectively improve the accuracy of summarization and can be applied to automatic text summarization.