BERT-Driven Deep Learning Approach for Depression Detection in Social Media Posts

Depression constitutes a worldwide health emergency, impacting countless individuals daily. However, timely diagnosis and treatment for depression are infrequent. Recent research has revealed affirmative links between the use of social media and the presence of depression. The use of learning techniques has become increasingly popular in detecting signs of depression in social media posts. This study introduces a method that utilizes transformer based models like BERT and RoBERTa. During the methodology phase we carefully preprocess the data followed by word embedding procedures. We then employ a combination of deep learning architectures such as LSTM, CNN with BiLSTM, RNN, BERT with Dense and RoBERTa with Dense to thoroughly evaluate and validate our models performance. Interestingly our results consistently show that the BERT with Dense and RoBERTa with Dense approaches outperform methods achieving an accuracy rate of 98%. This refined approach allows for accurate identification of signals in social media posts offering promising potential to make significant contributions, to the field of mental health advancement.