Depression Detection in Bangla: A Hybrid Model with FastText and XLM-RoBERTa

Depression is a major mental illness that affects millions of people globally, with significant treatment gaps, especially in low-resource settings. Social media is a valuable source for the early detection of mental health conditions, including depression. However, effective sentiment analysis models for low-resource languages like Bangla are scarce due to limited annotated datasets. A novel hybrid model is presented in this study for detecting depressive text from Bangla social media. To address the lack of Bangla datasets, a new dataset of 16,081 text data was created by collecting and annotating Bangla tweets using FastText and XLM-RoBERTa. This dataset provides a crucial foundation for model training in Bangla, a low-resource language. The proposed model feeds FastText embeddings into XLM-RoBERTa to classify posts as depressive or non-depressive, achieving 90% accuracy, surpassing existing approaches. This hybrid model offers a powerful tool for mental health monitoring and early intervention in Bangla.