Machine Learning - eBook (EN)

Enabling faster suicide intervention among veterans Since 2012, RallyPoint, a social media platform designed for the broader US military community, has provided an online user experience focused on military service members, veterans, families, caregivers, and survivors to help them lead more successful and fulfilling lives. Among the millions of public discussions on the platform, a small percentage comes from members who share thoughts and behaviors about self-harm. The US Department of Veterans Affairs estimates that approximately 17 military veterans die by suicide each day—and RallyPoint has made it a priority to offer critical mental health resources and support to these men and women when they need it. Developing a way to quickly and accurately sift through these high-risk public posts created by a small minority of RallyPoint users is a challenge. In order to speed the discovery of these at-risk public posts, RallyPoint turned to the Amazon Machine Learning Solutions Lab and researchers at Harvard's Nock Lab. The Amazon Machine Learning Solutions Lab worked closely with RallyPoint to develop a machine learning model using Amazon SageMaker that can quickly analyze public posts on the RallyPoint platform and help determine whether there is an indication of self-harm. With the help of this machine learning model, RallyPoint has been able to successfully flag concerning posts quickly and accurately while reducing the amount of manual review needed to enable a potentially lifesaving intervention. Ongoing, RallyPoint and Harvard will continue to further refine the model while evaluating the best content (e.g., mental health programs, hotlines, support groups) and preferred method to surface information to users. In the long term, the goal of the solution will be to augment the community engagement by RallyPoint member administrators that takes place on the platform today when the risk of self-harm in this audience is identified. 18 18

• Cover