Emotional and Linguistic Cues of Depression from Social Media

Abstract

Health outcomes in modern society are often shaped by peer interactions. Increasingly, a significant fraction of such interactions happen online and can have an impact on various mental health and behavioral health outcomes. Guided by appropriate social and psychological research, we conduct an observational study to understand the interactions between clinically depressed users and their ego-network when contrasted with a differential control group of normal users and their ego-network. Specifically, we examine if one can identify relevant linguistic and emotional signals from social media exchanges to detect symptomatic cues of depression. We observe significant deviations in the behavior of depressed users from the control group. Reduced and nocturnal online activity patterns, reduced active and passive network participation, increase in negative sentiment or emotion, distinct linguistic styles (e.g. self-focused pronoun usage), highly clustered and tightly-knit neighborhood structure, and little to no exchange of influence between depressed users and their ego-network over time are some of the observed characteristics. Based on our observations, we then describe an approach to extract relevant features and show that building a classifier to predict depression based on such features can achieve an F-score of 90%.