A Research on Generative Adversarial Networks Applied to Text Generation

Using deep learning methods to generate text, a sequence-to-sequence model is typically used. This kind of models is very effective in dealing with tasks that have a strong correspondence between input and output, such as machine translation. Generative Adversarial Networks(GAN) is a generation model that has been proposed in recent years, which has achieved good results in generating continuous and divisible data such as images. This paper proposes an improved model based on GAN, specifically using the transformer network structure instead of the original general Convolutional Neural Network or Recurrent Neural Networks as generator, and using the reinforcement learning algorithm Actor-Critic to improve the model training method. By comparing experiments, and selecting the perplexity, the BLEU score, and the percentages of unique n-gram to evaluate the quality of the generated sentences. The results show that the improved model proposed in this paper perform better than comparative models on above three evaluation indexes. This verifies its effectiveness in text generation.