WGAN method based on text generation

Abstract

The invention, which belongs to the field of the deep learning neural network, discloses a WGAN method based on text generation. The method comprises the following steps: S1, constructing a Wasserstein generative WGAN model; S2, preparing a text data set for training; S3, coding the test data by using a coder; S4, constructing random noises and acquiring obtain an output text of a generator; and S5, inputting the output text of the generator and the text after coding of the data set into a discriminator and carrying out network training. According to the invention, on the bass of characteristics of the generative WGAN model, a construction mode of combination of a deep convolution neural network and a WGAN is put forward creatively and the adversarial network training mode of the generatedtext is disclosed for the first time, so that the function that the previous generative WGAN model completes image generation is broken.

Description

technical field [0001] The invention relates to the technical field of deep learning neural networks, in particular to a WGAN method based on text generation. Background technique [0002] Generative Adversarial Network (GAN for short) is a framework proposed by Goodfellow in 2014. It is based on the idea of ​​"game theory" and constructs two models of generator (generator) and discriminator (discriminator). Uniform noise of (0, 1) or Gaussian random noise generates images, and the latter discriminates the input image to determine whether it is an image from the dataset or an image produced by the generator. Every time the discriminator completes a judgment, it returns the result error to the generator. [0003] However, in the traditional GAN ​​model, most of them let the generative confrontation network model complete the function of generating images, and there is no confrontation network training method involving text generation. Contents of the invention [0004] Th...

AI Technical Summary

Problems solved by technology

[0003] However, in the traditional GAN ​​model, most of them let the generative confrontation network model complete the function of generating images, and there is no confrontation network training method involving generated text.

Images