Abstract
<p>Depression is a psychiatric disorder that is highly prevalent worldwide, yet people often do not consult a clinician during its onset. They may instead convey their experience on social media platforms or in web-based forums. The development of automatic methods for detecting depression in these online spaces is growing. However, methods for gauging the severity of depression and the severity of different depression symptoms are less well explored. Such methods may improve our ability to detect severe depression early in users of these online spaces. In this study, we used datasets that contain the posts of Reddit users. Each user provided a completed BDI-II questionnaire, which revealed the severity of their depression and the severity of different depression symptoms. We extracted linguistic features from the users' posts, including first-person singular pronouns, negative emotion words, and absolutist words. We then performed regression analyses to understand the relationships between these linguistic features and the severity of depression and its symptoms. Our findings showed significant associations with the overall severity of depression, particularly with regard to negative emotion words. These results may have uses to improve methods for detecting early signs of depression and gauge its severity in the absence of formal tests. Our analysis did not find any associations between language features and fine-grained metrics of severity for individual depression symptoms. As these associations are not implausible, larger datasets may be needed to improve our linguistic understanding of different symptoms and symptom domains.</p>