Depression Detection on Social Media with Reinforcement Learning

Depression detection is a significant issue for human well-being. Conventional diagnosis of depression requires a face-to-face conversation with a doctor, which limits the likelihood of the identification of potential patients. We instead explore the potential of using only the textual information to detect depression based on the content users posted on social media sites. Since users may post a variety of different kinds of content, only a small number of posts are relevant to the signs and symptoms of depression. We propose the use of reinforcement learning method to automatically select the indicator posts from the historical posts of users. Our experimental results demonstrate that the proposed method outperforms both feature-based and neural network-based methods (over 14.6% error reduction). In addition, a series of experiments demonstrate that our model can deal with the noise of data effectively and can generalize to more complex situations.