Depression Detection Based on Multilevel Semantic Features

Depression is a common mental health disorder that can affect a person's mood, thoughts, and behavior. In this paper, we propose a depression detection method based on multilevel semantic features. This method consists of a character semantic feature extraction module, a keyword semantic feature extraction module, and a global semantic vector, which extract depression features from different perspectives in the text. Meanwhile, an Inception module is introduced into TextCNN to capture feature information at different scales. To evaluate the effectiveness of this method, we collected a dataset from a psychiatric hospital specializing in mental disorders, including symptom descriptions of depression patients. We conducted experiments on this dataset and the publicly available dataset CMDC, comparing our method with mainstream depression detection algorithms. Our method achieved accuracies of 0.97 and 0.94 on these two datasets, respectively, demonstrating that our method can effectively identify depression patients.