Development in social network communicationencourages dangerous use. An increasing number of mental disorders in socialnetworks, such as cybernetic dependency, information overload and networkconstraint, have recently been observed. Currently, the symptoms of thesemental illness are passively observed, which causes late clinical intervention.In this paper, we argue that online social behaviour mining offers the chanceto effectively recognize mental disorder at a beginning period. It is difficultto detect disorder because the mental state cannot be observed directly fromthe registers of online social activities. Our new and innovative approach tothe act of disorder detection is not based on self-disclosure of these mentalfactors through psychology questionnaires. Instead, we propose a framework ofmachine learning or the detection of mental disorders in social networks, whichexploits the features extracted from social network data to accurately identifypossible disorder cases. We also use multiple learning sources in socialnetwork mental disorder and propose a new mental disorder based model toimprove accuracy. Our framework is evaluated through a user study with none ofthe users on the network. We performed an analysis of the characteristics andwe also applied the new proposed approach in large-scale data series andanalysed the characteristics of the three types of mental disorder.