An analysis of mental health of social media users using unsupervised approach

With the shift of population toward digital lifestyle, it is becoming increasingly far easier to express opinions, behavior and mindset online - instantly and openly due to majorly following three very important reasons: firstly the online disinhibition effect/anonymity, second is the psychological distance and the third the emotional contagion (Lieberman and Schroeder, 2020). The myriads of data originating from social media platforms provide a major insight into the life of people. This insight underlines the mental health and emotional conditions of the users, chiefly the young population. The challenge is to identify the users who are showing signs of succumbing to mental illness at its onset (prodrome period). In this proposed method, we applied unsupervised algorithms on the data, signaling behavior change for psychological analysis and identified the probability of users showing at-risk behavior. By at-risk behavior, we mean users who are on the verge of acquiring some mental illness. In this study, we analyzed posts and tweets from the social media platform namely Twitter and developed an unsupervised model to classify users based on the scale of change in their behavior. Our model has achieved 76.12% accuracy.