The LDA topic model is a document generation model, which is an unsupervised machine learning technique that can be used in keyword extraction, topic classification, and so on. The main purpose of this paper is to effectively evaluate the number of optimal topics of the LDA topic model in the message subject classification and extract the importance of the topic, so that the LDA topic model can be highly readable after subject classification. Therefore, this paper proposes a keyword matching and subjective statistical value word comparison (KM-SSVW) for the subject classification of emails, the keyword matching technique in this method uses the TF-IDF technique. This method can accurately evaluate the extracted keywords and optimize the number of topics. The data in this article is mainly from the mail gate, a total of 7,000 messages that Hillary communicates with others. The empirical results show that the proposed method of keyword matching and subjective statistical value word comparison has a good effect on subject quantity optimization and subject word readable evaluation. However, there are still some limitations in this paper. In the experiment, new methods are not validated for other types of data sets, such as microblog short text, XML documents, and WeChat public platform articles.