Effective social dialogue patterns must understand rhetorical means of expression in a broad sense. Rhetorical questioning is a commonly used rhetorical expression that typically employs an interrogative form to convey a negative function. As the analysis progresses, we find that there is a continuum from interrogative to rhetorical question, with the degree of questioning decreasing and the degree of negation increasing. The rhetorical question allows for a response, and the form of the response reflects the addressee’s understanding of the degree of negation. The article manually marked 1002 rhetorical questions by experts and divided them into four categories: Awakening, Questioning, Discovering, and Constantly Changing. By comparing CNN, RNN, Transformer, FastText, Baidu PaddlePaddle model, and CNN with FastText composite model, it was found that CNN alone can achieve an accuracy of 53.12%, with good prediction performance, and performs well in various types of rhetorical questions. Therefore, CNN can be applied in natural language processing of rhetorical questions in fields such as sentiment analysis, translation, and automatic writing.