Depression and Suicide Risk Detection on Social Media using fastText Embedding and XGBoost Classifier

Abstract

Nowadays, people with depression reveal their mental conditions on social webs for emotional relief. Detection of suicide risk and depression content can save society and the young generation. A person at risk needs immediate medical attention, so early detection of depressive content with NLP is an important research area. We propose a novel framework to differentiate between depression and suicidal risk content with the fastText embedding for contextual analysis, TF-IDF vector for the relevance of terms, and machine learning classifier XGBoost for accurate classification. This novel approach achieves 0.78 AUC and 0.71 weighted F1-score for this Reddit dataset and also increases accuracy, weighted F1- score against one baseline model. Our experiments and analysis exhibit strong performance against various embedding models and classifiers for this challenging problem.