Abstract: Microblog review is characterized by its easy publication, rapid dissemination, extensive influence, which brings the users great convenience but also attracts a large number of spammers' attention, making it a spam and bad information release platform. Therefore, the microblog comments must be detected and filtered effectively in advance. By use of the J48 decision tree classification to sort out the commonly used vocabulary in micro-blog spam, and the selection of microblog comment information of wider concern as the experimental training set, the reliability of the method chosen is verified based on the accuracy, recall rate and precision ratio of data, and the effectiveness of the proposed method is proved by the experimental results.