The online social network has become a favorable site where a large number of malicious netizens spread rumors and conduct malignant competition. In this paper, we set up a method for generating review content in specific domain of social networks, which uses a recurrent neural network model to generate the social network-style review. Taking Twitter platform as an example platform, we firstly classify the review text according to the sentence pattern; secondly, aiming at different categories, we design corresponding recurrent neural network model to generate the initial review text corresponding to sentence structures; finally, we conduct automatic replacement of the generated initial text through the relevance of subject terms to achieve the effect of better adapting to hot topics. This method is not only easy to operate and economical, but also can evade the most advanced detectors. In the same environment, it is superior to the existing technology and generates more than 85.2% of the output text with correct grammar and wise contents.