Traditional context-aware systems on mobile platform mainly focus on utilizing various localization based technologies to detect and recognize significantly meaningful places. However, they cannot intuitively describe the dynamic semantic context of the surroundings. In this paper, a novel context sensing approach is proposed to distinguish typical context based on dynamic Bluetooth information. The study builts a context classification model through observing the occurrence of ambient Bluetooth devices and dynamic statistical features extraction and further applied the model into inferring semantic social context based on Bluetooth traces from real-world personal lives. Evaluation results show, just based on dynamic Bluetooth information, the proposed feature extraction methods and DT (Decision Tree) can achieve an average accuracy of 86.8% for recognizing six representative short time-length contexts, which outperforms several traditional machine learning methods. In addition, the accuracy of long time-length context inferring can also reach 92% without any additional information but Bluetooth.