A Deep Learning-Based Framework for Depression Detection in Digital Health Care Systems

In the era of rapid technological advancement, Twitter/X has emerged as one of the most widely used microblogging platforms. Users increasingly express their emotions, opinions, and personal experiences online, generating large volumes of textual data that can serve as valuable resources for analyzing mental health conditions. In recent years, depression has become a major public health concern worldwide, with the number of affected individuals increasing daily. In this study, we investigate the use of several deep learning models for the automatic detection of depression from Twitter/X data. The evaluated models include Long Short-Term Memory (LSTM), Gated Recurrent Units (GRU), Bidirectional Long Short-Term Memory (BiLSTM), Bidirectional Encoder Representations from Transformers (BERT), and a baseline Feed-Forward Neural Network (FFNN). In addition, we propose a hybrid architecture that combines BERT, GRU, and BiLSTM in order to capture both contextual semantic information and sequential linguistic patterns within tweets. The experimental results demonstrate that the proposed hybrid BERT-GRU-BiLSTM model outperforms the other models, achieving an accuracy of 94%, which highlights its effectiveness and efficiency in automatically detecting depression from Twitter/X data.