Bangla Depressive Social Media Text Detection Using Hybrid Deep Learning Approach

Tapotosh Ghosh, M. Shamim Kaiser

Abstract: Depression is considered as one of the major reasons of suicide which also affects personal and professional life of an individual. As people nowadays prefer to use social media, analysis of the contents generated by a person can lead us to get an insight of his mental health. For this, automated systems are essential due to huge explosion of social-media-generated data and privacy issue. In this paper, a deep learning approach based on Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Convolutional Neural Network (CNN), and Word2Vec embedding has been proposed to recognize depressive Bangla social media texts, which obtained 92.25% accuracy, 94.46% sensitivity, and 91.15% specificity. A dataset of 4000 Bangla social media posts was also created. The proposed approach was found better performing than LSTM, GRU, and classical machine learning models that proved its effectiveness in Bangla depressive text detection.