Detecção de comportamento depressivo com base em evidências textuais

Abstract

Depression is a condition that has become commonplace these days, with estimates of 300 million people around the world suffering from some variation of depression (WHO, 2018). Among all people affected by this condition, a large portion do not know they have depression, as many of these cases are undiagnosed (CEPOIU, 2008). There are several reasons that can make this diagnosis difficult, such as social stigma regarding treatment, lack of access to a qualified professional, given the cost of a therapy session, or even the need for initiative of the individual affected by this condition to seek a treatment. Although there are instruments, such as questionnaires, interviews, etc, which aim to assess the existence of this condition, they still have the same limitation cited regarding the accessibility. There are works in the area of computing proposing models that aim to alleviate this problem, using data from social networks of these individuals and trying to show patterns that characterize the behavior of an individual affected by this condition. Although these works produce an interesting result regarding the identification of an individual with depressive behavior, they do not explicitly present the psychological concepts involved, which guided the choice of characteristics used to describe depressive behavior. This work proposes a computational model of depression that partially addresses a psycho-linguistic model. A discussion is also presented regarding the qualities and limitations of the computational representation of some characteristics used by the psycho-linguistic model. At the end, the application of the proposed model on a real world data set is presented. The classifier had an accuracy rate of 73 % and a recall rate of 80 %.