CHS: Small: Collaborative Research: Tools for Mental Health Reflection: Integrating Social Media with Human-Centered Machine Learning

Georgia Tech Research Corporation, Atlanta GA

Investigators

Abstract

The widespread adoption of wearable devices and social media is generating population-scale data about people's behavior as situated in their everyday lives. Prior research has shown how machine learning techniques can use such data in modeling attributes of individuals' health and wellbeing; these techniques are also promising additions to tools that support self monitoring practices and health interventions outside of clinical contexts. However, most such tools enable only simple mechanisms to review one's data, and require high compliance from individuals actively volunteering relevant information. These limitations prevent such tools from effectively supporting reflection, that is, conscious re-examination of prior experiences to form new understanding. Reflection is a key to improved health and health maintenance, and recent work has shown the promise of data-driven health reflection. Effectively supporting reflection, however, requires more sophistication than simply showing a patient their data. This project will develop tools to support reflection for eating disorders (ED) by combining voluntarily shared and unobtrusively gathered social media data with strategic presentation of machine learning analyses. The interface designs will meet the needs of multiple stakeholders: patients, family members, and clinical partners. By doing so, the research will result in novel mechanisms to support the treatment of ED, going beyond existing personal health informatics tools by being sensitive to the complex psychological struggles of ED patients. The proposed research will follow a multi-phase process, interleaving the use of machine learning and human-centered approaches. The first phase will seek to understand the current practices of three stakeholders, patients, clinicians, and support network members, involved in ED reflection. In the second phase, informed by those current practices, we will develop theoretically-motivated, and psychometrically and clinically validated, machine learning techniques to support ED inference and reflection based on analysis of both textual and visual social media data. The third phase will use participatory design methods to develop interactive tools that encapsulate these machine learning techniques in order to support ED reflection among the three stakeholders. This final phase will include evaluating these tools through a field deployment to understand how they become embedded in and affect current practices of ED reflection. These activities will lead to complementary contributions in the areas of machine learning and of designing for reflection, offering novel approaches to outstanding challenges surrounding mental health and facilitating novel collaborations between computational and clinical researchers. Broader implications of the research will include conducting mental health outreach activities in the researchers' respective campuses and facilitating the training of the next generation of cyber-human researchers and professionals. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.