Nowadays SinaWeibo has become one of the most popular social media platforms both at home and abroad. However, the open and anonymous environment of this type of platforms provides rumors a perfect hotbed to breed and spread, and the negative influence on society from rumors cannot be ignored. Traditional rumor detection methods based on features often focus on static flat features of message contents, users, propagation and so on, but the  information propagation structure and the reaction of the propagation group are not fully studied. Aiming at this problem, first we introduce the cascade model of information propagation into the labeled propagation tree (LPT) and propose an improved modelLabeled Cascade Propagation Tree (CALPT). Secondly, we investigate users’ influence assessment by a dynamic method. Finally, we predict whether a microblog post is a rumor by applying 10 flat features with new features and hybrid kernel SVM based on random walk graph kernel and RBF kernel. Extensive experiments on realworld data from SinaWeibo demonstrate that the proposed method can improve the performance of rumor detection.