Emotional analysis of microblog can discover that the public's attitude towards hot events can grasp the network public opinion, so it has become a hot research topic in text mining. In view of the current situation that most of the existing affective analysis methods separate the deep learning model from the emotional symbols, this paper proposes a microblog affective analysis method based on dual attention model. This method uses Word2Vec tool to express microblog platform and build emotional dictionary; uses deep belief network and attention model to construct microblog emotional classification model; and trains the emotional classification model based on the constructed semantic representation. By combining attention model with emotional symbols, this method effectively enhances the ability of capturing emotional semantics of microblog text and improves the performance of microblog emotional classification. Based on Spark cluster, the deep confidence neural network is processed in parallel. The experimental results show that the model achieves the best results in three indicators. The recall and precision of the model are more than 0.04 higher than the traditional shallow learning method. INDEX TERMS Emotional analysis algorithms, deep learning, deep belief network (DBN), spark parallelization, text mining.