Depression Detection from Bangla Text Using BanglaBERT and Hybrid Deep Learning Techniques

Emotions such as depression, sadness, happiness are internal feelings that expose the actual behavior of a person. These type of emotions are very important and sometimes it can be difficult to examine them properly. Among them, depression has become a common mental disorder that can affect our mental health and daily activities. Social media like Facebook, Twitter, YouTube are getting a part of the daily activities of people. They often use these platforms to share their personal experiences, beliefs, opinions etc. In this study, we have applied BanglaBERT, CNN-BiLSTM (Convolutional Neural Network with Bidirectional Long Short-Term Memory), BiGRU-CNN (Bidirectional Gated Recurrent Units with Convolutional Neural Networks), and GRU-LSTM (Gated Recurrent Units with Long Short-Term Memory) to detect depression from the Bangla comments of YouTube videos. To accomplish the research, we have utilized three different datasets containing Bangla textual data. Besides, we have applied hyperparameter tuning method to train the models with the best possible parameters. As a result, we have found that the BanglaBERT model has outperformed the CNN-BiLSTM, BiGRU-CNN and GRU-LSTM hybrid models overall in all three datasets in terms of Accuracy, Precision, Recall and F1-Score.