Depression Detection on e-Risk 2017 using Long Short-term Memory Models

In today’s fast-paced world, individuals are increasingly vulnerable to high levels of stress, which consequently raise the risk of developing depression. Depres-sion may represent itself via various symptoms, ranging from emotional to physical pains. The symptoms appear gradually, deceiving patients to ignore their changes. The reluctance to accept the depression and the social stigma surrounding this mental disorder prevent a remarkable fraction of patients from vising the specialists and receiv- ing the treatments. Although people feel uncomfortable to reveal themselves in face-to-face connections, they have an opposite attitude towards social networks. People willingly share their opinions, thoughts, and feelings on social networks nowadays. Psychological studies have explored potential differences in the literature used by individuals with depression and those without. Hence, social network data has become a valuable source of information to study the mental health of users. Following this motivation, in this paper, we have examined the power of several deep learning algorithms for automati- cally detecting depression on eRisk 2017 dataset, obtained from the social network of Reddit. To this aim, we have studied and compared Long Short-Term Memory (LSTM), Bidirectional LSTM (Bi-LSTM), Gated Recurrent Unit (GRU), Bidirectional GRU (Bi-GRU), and hybrid LSTM- GRU models for depression detection. The results have illustrated superiority of LSTMs in terms of accuracy and timing of detecting depression over the rest of the models. These findings contribute to the expanding body of research on using artificial intelligence for detecting mental health issues.