Vital node detection in a social graph is a long-standing challenge for understanding the exchange, sharing, and dissemination of information on social media. As one of the most popular social media platforms in China, the Sina Weibo microblogging website has millions of users. Of these, vital users play an important role in impacting other common users’ online behaviors, promoting online information dissemination, and leading public opinion trends. Therefore, it is of theoretical and practical interest to study the identification of vital users on the Sina Weibo network. In contrast to traditional approaches, after investigating the unique characteristics of the vital users in the Sina Weibo network, we proposed an end-to-end computational model based on the deep graph neural network. The proposed model introduced a simplified graph attention mechanism to incorporate the information of both the follower and followee networks of users, which can maintain a relatively low computational cost compared with the traditional GAT model but at the same time maintain comparable detection accuracy. The comparison experiments between the proposed model, classic node ranking methods, and graph embedding-based classifiers verified that the proposed model has the best accuracy in detecting vital users in the Sina Weibo network.