With the increase in Weibo users, the Weibo platform produces large amounts of data every day. The rapid propagation of data would result in an extensive change in public opinion. Thus, correctly predicting public opinion from Weibo is an urgent challenge. For this purpose, an approach to public opinion prediction from Weibo using an improved BP is proposed in this paper. First, according to the characteristics of Weibo public opinion, this paper constructs nine public opinion indexes to analyze Weibo public opinion. Third, because the BP is susceptible to the choice of initial weights and has a poor rate of convergence, GA is introduced to optimize the BP. However, the GA easily gets stuck in local optimal solutions. Therefore, the Metropolis acceptance criterion is employed to improve the local searching ability of GA. Then, the IGABP algorithm, which is based on the improved GA, is proposed. Finally, from extracting and normalizing the Weibo data, the validity of the IGABP algorithm is verified. The experimental result shows that the IGABP algorithm is feasible in the prediction of Weibo public opinion. In addition, the IGABP algorithm has better generalization ability and higher accuracy in Weibo public opinion prediction.