Abstract: In view of the difficulty in extracting features of Weibo's short texts and the existence of a large number of officially certified microblogs that had not been used efficiently, a microblog rumors detection method was proposed based on topics and prevention model. Firstly, the official rumors were extracted and categorized according to the subject and were organized according to the user, spread frame and content characters, forming the subsets of official rumors based on a certain topic. And then, the similarity between the microblog and the official rumors with an identical topic was calculated. Merging the values with the traditional features, the result was taken as statistical features put into supervised machine learning. Finally, some experiments were carried out to validate the detection method. The results show that, compared with the traditional supervised machine learning, the method can improve the performance of Weibo rumors detection by about 3%, and can achieve rumor prevention.