Institution(s): Center for Artificial Intelligence in Society, University of Southern California
Author(s): Amulya Yadav, Eric Rice, Robin Petering, Jaih Craddock, Bryan Wilder, Milind Tambe
Homeless youth service providers implement social network based peer-leader intervention programs among homeless youth to help prevent HIV infection. In these interventions, a select number of youth, called peer leaders, are taught about how to change their behaviors to reduce the chances of contracting HIV. These leaders are then encouraged to share these messages among their peers in their social circles. We developed HEALER, a decision support system which assists social workers in selecting the most "influential" peer leaders for their social network based interventions. First, HEALER relies on online contacts and friendship based information of homeless youth (provided by social workers) to create a social network of the youth and their connections. This information is then analyzed by HEAL, an algorithm which utilizes state-of-the-art AI techniques from sequential decision making under uncertainty and decision theory, to pinpoint which homeless youth in the network would make successful peer leaders. The social workers then educate these peer leaders about HIV prevention, and encourage them to share their knowledge in their social circles. Finally, social workers are able to gather more data about the network based on feedback from the peer leaders. This information is passed back to HEALER, which enables it to continually refine its results for future interventions. HEALER has proven to be effective in the real-world. Results from real-world pilot studies show that HEALER outperformed current modus operandi of conducting network based interventions by almost 50%. This demonstrates that HEALER brings significant improvement over current approaches to network-based HIV interventions.