HAN-DistilBERT for Depression Detection from Twitter Data: Integrating Hierarchical Attention with Transformer-Based Model

Depression, a severe mental health disorder often linked to suicide, can be detected through social media content. However, analyzing depression-related signs is challenging due to the informal, brief, and varied nature of posts. To tackle this, this study presents a Hierarchical Attention Network (HAN) combined with DistilBERT to identify depression indicators in social media posts. Through the utilization of Hierarchical Attention Network (HAN) architecture that incorporates word-level attention using DistilBERT and sentence-level attention through a customized layer, this model significantly enhances its capability to distinguish between depressed and non-depressed posts with improved accuracy. Performance metrics such as accuracy, recall, precision, and F1 score are used to evaluate the model and the results demonstrate an outstanding accuracy of 96%. The integration of hierarchical attention with transformer-based embeddings has proven to make this model a powerful and reliable solution for detecting depression in social media posts.