Transforming Mental Health Monitoring: Bidirectional Encoder Representations from Transformers (BERT)-Driven Analysis of Social Media for Depression Detection

Abstract

The recent increase in the use of social media as a platform for expressing users' feelings and mental state necessitates the ability to detect signs of depression and anxiety from the user generated contents for early detection and support. However, this becomes quite a challenge due to the multilingual nature of global social media, and the development of mental health detection system with linguistic universality. The present work introduces a deep learning hybrid approach by combining mBERT (Multilingual Bidirectional Encoder Representations from Transformers) with BiLSTM (Bidirectional Long Short-Term Memory) network. The deep semantic context of text is captured using mBERT and its ability to learn sequential emotion pattern is further exploited by the BiLSTM layer. This model is further fine-tuned using annotated datasets with social media posts in different Indian languages, to classify multiclass (depression, anxiety, or neutral). The system is designed to capture linguistic and symbolic elements of texts including emojis and code-mixed content, which have high performance on all the linguistic datasets. This paper presents a robust, scalable hybrid framework for multilingual mental health detection, advancing the development of inclusive AI-assisted mental wellness monitoring tools.