Abstract: Depression is a major contributor to the overall global burden of disease. Traditionally, doctors have diagnosed depression in people face-to-face using criteria for clinical depression. However, more than 70% of patients would not consult a doctor in the early stages of depression, which leads to further worsening of their condition. Meanwhile, people increasingly rely on social media to reveal emotions and share their daily lives, so social media has been successfully used to help detect physical and mental illnesses. Based on these inspirations, our work focuses on the early detection of depression through social media data collection. We create a well-labelled depression and non-depression dataset on Twitter and extract six clusters of depression-related features that cover not only criteria for clinical depression but also online social media behavior. With these feature groups, we propose a multimodal depression dictionary learning model to detect depressed users on Twitter. Finally, we analyse a large dataset on Twitter to uncover underlying online behaviors between depressed and non-depressed users.

Keywords: Depression Detection , Machine learning, Twitter dataset.

| DOI: 10.17148/IARJSET.2023.10594