<p>The rapid development of internet technology has brought with it an increasing demand for privacy protection and data security. Analyzing encrypted traffic generated by social media user behavior can assist network administrators in resource allocation and the identification of malicious activities. Existing methods for classifying encrypted traffic suffer from issues such as unstable feature extraction leading to insufficient recognition accuracy and inadequate model generalization capabilities. To solve the above problems, in this paper, we propose UBI-STA, an identification method based on spatiotemporal correlation analysis. Firstly, we use a time series segmentation algorithm to represent encrypted traffic in segments, overcoming fluctuations in the network environment and extracting the temporal relationship of traffic data. Then, we design two deep learning classifiers based on hyperbolic space mapping (HB-CNN and HB-LSTM), introducing the semantic features of user behavior encrypted traffic into hyperbolic vector space, enriching the spatial semantic information of traffic data, and extracting spatial features of user behavior encrypted traffic. Finally, we utilize classification algorithms to achieve user behavior recognition. The results of extended experiments on a private dataset and a public dataset show that our method can effectively identify encrypted traffic generated by social media user behavior, with accuracy, precision, recall, and F1-score all superior to state-of-the-art methods.</p>