A two-stage text-to-emotion depressive disorder screening assistance from online community text

Abstract

Major Depressive Disorder (MDD), is one of the most serious mental diseases that can occur in anyone regardless of age or gender. Even though the symptoms of MDD are similar to regular emotional sadness, its impact is much worse and cloud lead to committing suicide. Nowadays, almost all people regularly use social media to communicate and express their opinions where the emotion could be impliedly illustrated through the textual conversation which leads to the implication to the possible sign of depression. This research presents a two-stage predictive model for MDD's risk screening assistance by using emotion weight based on textual data from social community. The messages are identified in six emotional dimensions angry, bored, excited, scared, happy, and sad. These emotions are then categorized into 6 emotion-based clusters using DBSCAN algorithm and examined by medical and psychological experts with Patient Health Questionnaire-2. The labeled clusters are used to train and optimize classification models for each screening question with several algorithms Decision Tree, Random Forest, and Gradient Boosted Tree. The experimental results show that the Gradient Boosted Tree is the best model for Q1 and the Random Forest is the best model for Q2 with 98.32% and 99.98%, accuracy respectively. This research will be beneficial in the further study in identifying depression from text-based emotion.