With the explosive development of the Internet, information and its spread across the network have grown exponentially. The rapid growth of information places higher demands on the information-censoring work of governments, even in areas where government censorship is rigorous. Astroturfing in political science is the practice of a government or an organization masking ideas with specific content designed to influence ordinary individuals or manipulate the orientation of public opinion. When we try to analyze social media content for multiple tasks, it is difficult to distinguish this government propaganda content from the views of ordinary citizens. In this thesis, a deep learning-based method proposed for detecting the Fifty Cent Army was trained with a database, named CSMCDB, of 24,788 labeled comments posted on mainstream Chinese social media and news websites. According to multiple test cases, this method allows users to predict Fifty Cent Army comments from different resources with high accuracy.