Mental Disorder Indication Detection with Transformer Model Architecture

Mental disorders are conditions that greatly affect a person's mindset, emotions, and behavior. The impact will greatly affect mood, thoughts, feelings, behavior, despair, or the most severe is suicidal tendencies. Some types of mental disorders that are often experienced include depression, anxiety, or even bipolar and schizophrenia. Every year, the number of people with mental disorders continues to increase. In 2022, there were 970 million people worldwide struggling with mental illness. However, there are limitations in providing health services for early detection of mental disorders. In addition, mental disorders are often considered taboo or even disgraceful, so most sufferers tend to be reluctant to talk about their mental health problems or tend to refuse to visit health services. Advances in technology and social media encourage the availability of freedom in storytelling. So, it is possible to develop a model for detecting indications of mental disorders using content on social media. Several Deep Learning models such as Long Short-Term Memory (LSTM) or Convolutional Neural Network (CNN) have shown good performance despite having limitations in understanding the overall context of words. Therefore, this study uses a Transformer-based model such as Bidirectional Encoder Representations from Transformers (BERT) to improve detection performance. A variation of this model is to add BERT architecture parameters to the LSTM or CNN model at the classification stage. The results of this study indicate that the addition of the BERT model can provide more optimal results, especially in the BERT CNN model with an accuracy of 96%.