Psychological stress is frightening the people’s fitness. It is important to notice stress suitable for active care. With the reputation of social media, people are reprocessed to sharing their day-to-day events and cooperating with friends on social media platforms, building it achievable to strength online social network data for stress detection. In this paper, we finding that users stress state are carefully associated toward that of friends in social media, and we pay a largescale dataset from actual social stands to thoroughly study the connection of users’ pressure situations and social communications. We major outline a set of stress-related documented, graphic, and social qualities from numerous features, and then suggest a novel hybrid model - a factor graph model joint with Convolution Neural Network to influence tweet content and social contact information for stress discovery. Investigational results display that the proposed model can recover the detection show by 6-9% in F1-score. By further analyzing the social interaction data, we also discover several fascinating marvels, i.e. the number of social structures of scarce influences (i.e. by no delta connections) of stressed users is around 14% progressive than that of non-stressed users, representative that the social structure of stressed users’ friends incline to be less connected and less complicated than that of non-stressed users.