Modeling Fine-Grained Dynamics of Mood at Scale ⇤

Mental health affects all aspects of people’s lives, yet it remains difficult to obtain accurate data about influential factors. This work investigates quantifying, infering, and predicting—via social media data—the day-to-day mental state of individuals. We develop a statistical model of the affective state (mood) of specific individuals with up to hourly temporal resolution. This model enables us to quantify, in a unified way, aggregate mood trends, as well as patterns specific to individuals and groups of friends. It finds key features of mood variation over time and allows us to decompose a person’s emotional state into a weighted sum of contributing factors—shedding new light on how mood affects, and is affected by environment. We then show that individuals’ mood can be accurately predicted days into the future based on online behavior.