NAT-CVAE: A Novel Method for Controlled Long Text Generation

In order to solve the problems of incomplete textual relational representations and low quality of model generation in the field of controllable long text generation, this paper proposes a new NAT-CVAE model (Neural network-based text graph structuring and Adversarial Learning, T-CVAE model). The model obtains a more comprehensive representation of text relations and then a richer potential representation by neural network-style graph structuring of the input text, and at the same time draws on the game idea of generative adversarial network to conduct adversarial training between the generated text and the target text, so as to improve the controllability and effectiveness of text generation. Experiments show that the model has excellent representation learning ability, and also achieves satisfactory results in text effectiveness and controllability, and is able to generate long texts that meet the given constraints.