Suicide is one of major public health problems worldwide. Traditionally, suicidal ideation is assessed by surveys or interviews, which lacks of a real-time assessment of personal mental state. Online social networks, with large amount of user-generated data, offer opportunities to gain insights of suicide assessment and prevention. In this paper, we explore potentiality to identify and monitor suicide expressed in microblog on social networks. First, we identify users who have committed suicide and collect millions of microblogs from social networks. Second, we build suicide psychological lexicon by psychological standards and word embedding technique. Third, by leveraging both language styles and online behaviors, we employ Topic Model and other machine learning algorithms to identify suicidal ideation. Our approach achieves the best results on topic-500, yielding F_1-measure of 80.0%, Precision of 87.1%, Recall of 73.9%, and Accuracy of 93.2%. Furthermore, a prototype system for monitoring suicidal ideation on several social networks is deployed.
展开▼
