Chinese text abstract generation method based on sequence-to-sequence model

Abstract

The invention discloses a Chinese text abstract generation method based on a sequence-to-sequence model, and the method comprises the steps: firstly carrying out the word segmentation of a text, filling the text to a fixed length, and carrying out the Gaussian random initialization of a word vector; encoding the text, inputting the encoded text into a bidirectional long short-term memory (LSTM) network, and taking the final output state as precoding; performing convolutional neural network (CNN) on the word vectors according to different window sizes, and outputting the word vectors as windowword vectors; constructing an encoder, constructing a bidirectional LSTM (Long Short Term Memory), taking precoding as an initialization parameter of the bidirectional LSTM, and taking a window word vector in the previous step as input; and constructing a decoder, and generating a text by using a one-way LSTM and combining an attention mechanism. According to the method, a traditional encoder froma sequence to a sequence model is improved, so that the model can obtain more original text information in an encoding stage, a better text abstract is finally decoded, and a word vector with smallerfine granularity is used, so that the method is more suitable for a Chinese text.

Description

technical field [0001] The invention relates to the technical fields of deep learning and natural language processing, in particular to a method for generating Chinese text abstracts based on a sequence-to-sequence model. Background technique [0002] The main content of the text automatic generation summary technology is as follows: for a long text, a short text is finally generated through the model, and the main content of the source text can be summarized. [0003] There are currently two mainstream text summarization methods, extractive and generative. The extractive method refers to using an algorithm to find one or several sentences closest to the main idea from the original text. It is a relatively mature solution, but since all the content of the extractive summary is extracted from the original text, the generated summary is readable. The performance and fluency are unsatisfactory, and there is still a long way to go from practical application. [0004] The gener...

AI Technical Summary

Problems solved by technology

[0008] The generative method based on the deep learning model will have a better performance in the final generated summary, but there will also be problems such as missing some important information and semantic incoherence. Most of the current improvement solutions start with the decoder and improve the decoding. way and adjust the attention mechanism, but the effect is still very limited, and the training time is longer

Images