Abstract
Depression detection in social media can be regarded as a complicated task, given the complex nature of mental disorders. Studies have shown that depression has an effect on language usage and many depressed individuals take on online platforms to self-disclose or discuss mental health issues, such as depression. Social media environments present an opportunity for researchers to potentially identify mental illnesses through the use of modern approaches, such as machine learning and neural networks. Existing solutions, however, rely mostly on the amount and quality of content posted by users and ignore key aspects of social networks - ego neighborhoods. We propose a novel method, MentalNet, for early detection of depression, based on Deep Graph Convolutional Neural Network (DGCNN), which considers the textual content of user’s posts, as well as the user relationships formed on social networks. The relationship data is based on user interactions which are captured using graph neural networks for better depression detection. Extensive experiments show that MentalNet outperforms state-of-the-art methods in depression detection, with up to 14%, 17%, and 19% improvement on precision, recall, and F1 score respectively.