DISCOVERY OF STRAIN IN ONLINE NETWORKS WITH THE SUPPORT OF PUBLIC INTERACTIONS

Psychological stress threatens people's health. It is not easy to spot stress in a timely and proactive manner. With popular social networking, people are accustomed to sharing their daily activities and interacting with friends on social media platforms, making it possible to take advantage of social network data over the Internet to detect stress. In this paper, users' stress status is closely related to their friends in social networks, and we use a broad set of data from real social platforms to conduct a systematic study of the relationship of stress. And social interactions of users. First, we define a set of textual, visual and social characteristics associated with stress from multiple aspects, and then propose a new hybrid model: a global graphical model shared with the neural network to leverage the use of Twitter content and social interaction information to detect stress. Experimental results show that the proposed model can improve detection performance by 5-10% in F1. In further analysis of social interaction data, we also found many interesting phenomena, i.e., the number of dispersed social connections (i.e., without Delta connections) of users is approximately 13% higher than non-fanatical users, indicating that the social structure of stressful friends tends to be less connected and less complex than non-authorized users