Verified account prediction on Sina Weibo with deep learning

The focus of the paper is to use a single weibo from a user to predict whether the user account is verified, referred to as verified account prediction, on Sina Weibo. To the best of our knowledge, verified account prediction on Sina Weibo has not been studied. For better understanding of the prediction problem, a comprehensive data analysis of weibos related to verified accounts is conducted first. Then, verified account prediction is formulated as a sequence learning problem. Specifically, a weibo from a user is represented as a sequence of feature values by feature hashing and whether the user account is verified is the corresponding label to predict. A deep learning approach is proposed for solving verified account prediction in this formulation. The proposed approach significantly outperforms the shallow learning methods in the comparisons in terms of accuracy and F1 by large margins in the experiments.