Abstract

Social media platforms are increasingly being used for study-ing psycho-linguistic phenomenon to model expressions ofsuicidal intent in tweets. Most recent work in suicidal ideationdetection doesn’t leverage contextual psychological cues. Inthis work, we hypothesize that the contextual information em-bedded in the form of historical activities of users and ho-mophily networks formed between like-minded individualsin Twitter can substantially improve existing techniques forautomated identification of suicidal tweets. This premise isextensively tested to yield state of the art results as comparedto linguistic only models, and the state-of-the-art model