Modeling and Understanding Visual Attributes of Mental Health Disclosures in Social Media

Content shared on social media platforms has been identified to be valuable in gaining insights into people's mental health experiences. Although there has been widespread adoption of photo-sharing platforms such as Instagram in recent years, the role of visual imagery as a mechanism of self-disclosure is less understood. We study the nature of visual attributes manifested in images relating to mental health disclosures on Instagram. Employing computer vision techniques on a corpus of thousands of posts, we extract and examine three visual attributes: visual features (e.g., color), themes, and emotions in images. Our findings indicate the use of imagery for unique self-disclosure needs, quantitatively and qualitatively distinct from those shared via the textual modality: expressions of emotional distress, calls for help, and explicit display of vulnerability. We discuss the relationship of our findings to literature in visual sociology, in mental health self disclosure, and implications for the design of health interventions.