Depression is a common illness that negatively affects how people feel, think, and act. It causes feelings of sadness and sleeping disorders. In serious cases, it leads to self-harm or suicide. Many researchers in computer science addressed the problem of depression detection. However, less research concerns the emotional upheaval of depressive people and investigates the reasons behind the depression. In this paper, a deep learning model is first constructed to automatically determine the negative sentiment degree for a Facebook post. The curves of emotional upheavals for depressive users are then generated. Based on the post contents, weather, and news data, relevant events are detected to infer the reasons of the negative emotions. A correlation analysis between the behavioral data of the depressive users on Facebook and their negative emotions is also conducted. The results of this study can not only provide a self-examination tool for depressive people, but also serve as a diagnostic assessment reference for medical personnel.