Harnessing Social Media to Identify Homeless Youth At-Risk of Substance Use

Homeless youth are a highly vulnerable population and report elevated rates of substance use. Prior work on mitigating use among homeless has primarily relied survey data to get information about youth, which can then be used inform the design targeted intervention programs. However, such is often onerous collect, limited by its reliance self-reports retrospective recall, quickly becomes dated. The advent social media provided us with an important source for understanding health behaviors youth. In this paper, we target specific demonstrate how detect based texts from media. We collect ~135K Facebook posts comments together responses group build novel detection systems machine learning natural language processing techniques. Experimental results show that our proposed methods achieve ROC-AUC scores ~0.77 identifying certain kinds using conversations only, ~0.83 when combined answers four questions not their demographic characteristics or Furthermore, investigate connections between people's provide insights problem.