Detecção automática de postagens possivelmente depressivas em redes sociais

Abstract

Currently, depression is one of the most worrisome mental health issues. In Brazil, in 2019, 10.2% of the adult population reported having been diagnosed with depression according to data from the National Health Survey. Identifying people with a possible depressive profile allows adequate monitoring by mental health professionals. In this sense, online social networks such as Twitter can be important allies. This monography presents experiments carried out for the automatic classification of Twitter posts (or a collection of posts produced by a given user) containing content that denotes some symptom of depression, as well as classification of depressive posts and users through an ensemble model composed of symptom classifiers. Logistic regression showed the best results in both symptom and depression classification tasks (average F 1 equal to 57% for the former, F 1 equal to 64% for the latter). This work is part of the Amive project (FAPESP Regular Grant #20/05157-9).