An Enhanced BERT Model for Depression Detection on Social Media Posts

Depression and other forms of mental illness are relatively common, and it has been shown that these conditions have an effect on an entity's physical health. Newly, artificial intelligence (AI) technologies have been created to aid mental health practitioners, such as psychiatrists and psychologists, in decision-making based on the historic data of patients (for example, medical records, behavioral data, social media use, etc.). These AI methods are intended to help mental health clinicians treat patients more effectively. One of the most recent generations of AI technologies, deep learning (DL), has exhibited greater performance in a wide variety of real-world applications spanning from computer vision to health care. When adopting bidirectional encoder representations from transformers (Enhanced BERT), the authors of the current research offer a new framework that may quickly and accurately identify postings that are connected to anxiety and depression. In addition, an intelligence distillation approach is a present method for transferring information from a large pretrained model BERT to a reduced model in instruction to expand the performance and accuracy of the smaller model. Researchers made use of word2vec and BERT in order to effectively analyze and identify symptoms of melancholy and anxiety based on our very own 40,000 data collecting infrastructure based on Twitter, the most widely used of the social media platforms. Using the Enhanced BERT methodology, our system achieves an accuracy of 92%, which is difficult than any other state-of-the-art technology.