Detection of Mental Illness Risk on Social Media through Multi-level SVMs

As shown by several previous studies, personal narratives through social media are often indicative of one's psychological state. In particular, mental illnesses such as depression were found to be associated with distinct linguistic patterns. However, many people with mental illness still do not receive full treatment. In this paper, we study mental illnesses through people's choice of words in expressing themselves on two popular social media platforms, Reddit and Twitter. Our goal is to develop an empirical model to detect and diagnose major mental disorders in individuals. We build a substantial dataset of posts made by people suffering from mental illnesses and the control ones, and in order to generate numerical feature from text we apply text cleaning and Word2Vec language modeling, and then for classification we used SVM machine which classifies posts and users with high accuracy. We achieve an accuracy of 95% on Twitter users and an accuracy of 73% on the Reddit challenge.