Abstract: Natural language generation is an important part of artificial intelligence system. As artificial intelligence technology is gradually integrated into human lives, people have higher requirements for natural language generation technology, and controllable natural language generation has become a new research hotspot. The task of text style transfer, as an important research direction of controllable natural language generation, has been widely concerned by academia and industry. The task of text style transfer refers to the purposeful generation of target-style sentences on the basis of retaining the original content of the text. To further demonstrate the development process of this task, the definition and development context of the current text style transfer task was first introduced. Then, based on the model method, the current research work was summarized into three categories: based on latent space disentangled, and based on explicit disentangled, based on one-step mapping, and on the basis of a brief description of these works, enlightening work was focused on. Furthermore, the common data sets and evaluation methods in the text style transfer task was combed and analyzed. Finally the future development trend of the text style transfer task was prospected.