Probability-Based Multi-Label Classification Considering Correlation Between Labels— Focusing on DSM-5 Depressive Disorder Diagnostic Criteria

Abstract

The incidence of depressive disorder in Korea is the highest among OECD countries. The proportion of patients in their 20s is the highest. However, social gaze and false perception are causing problems such as not visiting the hospital or delaying the visit. Accordingly, we suggest a Korean model for predicting depressive disorders using data from online communities widely used by people in their 20s. In many countries, including South Korea, depressive disorders are diagnosed using DSM-5 (Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition) published by the American Psychiatric Association. We propose a model that predicts the probability of a user’s speech corresponding to nine criteria for diagnosing DSM-5 depressive disorder, following advice obtained through periodic meetings with a psychiatrist. The prediction performance was improved by using the correlation between each criterion in the model implementation stage.

Keywords

• Artificial intelligence
• correlation
• depressive disorder
• multi-label classification
• natural language processing
• psychiatry in AI