Microblogging services such as Twitter and Plurk allow users to easily access and share different types of social multimedia (e.g. images and videos) over the online social world. However, information overload happens to users and prohibits them from reaching popular and important digital contents. This paper studies the problem of predicting the popularity of social multimedia which is embedded in short messages of microblogging social networks. Social multimedia exhibits the property that they might be persistently or periodically re-shared and thus their popularity might resurrect at some time and evolve over time. We exploit the idea of concept drift to capture this property. We formulate the problem using classification, and propose to tackle the tasks of Re-share classification and Popularity Score classification. Two categories of features are devised and extracted, including information diffusion and explicit multimedia meta information. We develop a concept drift-based popularity predictor, by ensembling multiple trained classifiers from social multimedia instances in different time intervals. The key lies in dynamically determining the ensemble weights of classifiers. Experiments conducted on the Plurk data show the high accuracy on the popularity classification and the promising results on detecting popular social multimedia.