摘要：Psychologists have used tests and carefully designed survey questions, such as Beck's Depression Inventory (BDI), to identify the presence of depression and to assess its severity level. On the other hand, methods for automatic depression detection have gained increasing interest since all the information available in social media, such as Twitter and Facebook, enables novel measurement based on language use. These methods learn to characterize depression through natural language use and have shown that, in fact, language usage can provide strong evidence in detecting depressive people. However, not much attention has been paid to measuring finer grain relationships between both aspects, such as how is connected the language usage with the severity level of depression. The present study is a first step towards that direction. We train a binary text classifier to detect ``depressed'' users and then we use its confidence value to estimate the user's clinical depression level. In order to do that, our system has to be able to fill the standard BDI depression questionnaire on users' behalf, based only on their posts in Reddit. Our proposal was publicly tested in the eRisk 2019 task obtaining the best and second-best performance among the other 13 submitted models.
