Predicting Signs of Depression from Twitter Messages

Abstract

Depression is a mental health problem that is experienced by many people around the world. Often, people with depression express their feelings via their posts on different social media platforms. If depression trace can be detected from their messages, it will help to understand their emotional states and to provide appropriate assistance. This paper proposes an application of natural language processing to this very important issue by predicting major signs of depression from Twitter messages in Thai. These major signs include Suicidal Ideation, Anhedonic, Sleep Problems, and Guilty Feelings. Different machine learning algorithms, i.e. Long Short-Term Memory (LSTM), Gated Recurrent Units (GRU), and Support Vector Machine (SVM) are used to build prediction models. A web application prototype is developed to predict signs of depression from a user's tweets during the past month to trace whether the user has shown any signs of depression as well as the degree or intensity of each sign in five scales. Such information can assist a mental health professional's client, who is experiencing depression, to realize one's own negative thoughts, and can be useful input to the treatment.