A Context-Aware Framework for Depression Detection in Multilingual Social Media Content

Depression is a mental health-related disorder and a global concern, yet many people remain undiagnosed due to a lack of awareness and limited access to mental health care. With the increasing use of social media for self-expression, there is a growing opportunity to detect signs of depression using automated systems. This study proposes a context-aware multilingual framework for detecting depression in social media text, focusing on both Bangla and English languages. Although the dataset primarily consists of Bangla social media posts, we converted emojis into their corresponding English textual meanings to enhance contextual understanding and introduce multilingual characteristics into the dataset. Three deep learning models—mBERT, Bangla-BERT, and BiLSTM—were implemented and evaluated on a well-curated, balanced dataset labeled as depressive or non-depressive. The results demonstrate that mBERT outperformed the other models, achieving an accuracy score of 90%, while Bangla-BERT and BiLSTM achieved accuracy of 88% and 86%, respectively. These findings highlight the effectiveness of transformer-based models, particularly multilingual ones, in enhancing the early detection of depression from user-generated content.