Optimization of Text Generation Method for Task-based Human-machine Dialogue System

The human-machine dialogue system is now used in all aspects of life, but the task-based system reply relies on manually marked rules and templates, which makes the text rigid and monotonous, and can not form coherent paragraphs when there is more information, but simply combine sentences together. In order to solve the above problems, this article first analyzes the construction method of the task-based dialogue system, and then proposes an optimized text generation method for the pipeline method. This paper uses the Planning-based Hierarchical Variational Model (PHVM) in the text generation phase. This model will make a global plan before generating the text, split the long text generation task into clauses, and retain the relevance between the clauses, and then generate clauses one by one. In addition, it uses sampling to introduce latent variables globally and locally to provide a variety of options for text generation. Experiments show that the model performs well in generating coherent text and rich text diversity, and solves the existing problems of task-based dialogue systems.