D$^4$: A Psychiatrist-proofread Dialogue Dataset for Depression Diagnosis

Anonymous

Abstract: Depression has affected large populations and become a significant threat to life expectations globally.Automatic depression diagnosis methods have been a new research focus. In particular, automatic dialogue-based diagnosis systems are desired since depression diagnosis highly relies on clinical consultation.Based on clinical diagnosis criteria, doctors initiate a conversation with ample emotional support that guides the patients to expose their symptoms. Such a dialog is a combination of task-oriented and chitchat, different from traditional single-purpose human-machine dialog systems. However, due to the social stigma associated with mental illness, the dialogue data related to the diagnosis of actual patients are rarely disclosed. The lack of data has become one of the major factors restricting the research on the consultation dialogue system of depression.%Although there are effective methods of diagnosis and treatment for depression, more than 75\% of people in low- and middle-income countries receive no treatment. Based on clinical depression diagnostic criteria ICD-11 and DSM-5, we construct a Psychiatrist-proofread Dialogue Dataset for Depression Diagnosis which simulates the dialogue between the doctor and the patient during the diagnosis of depression and provides diagnosis results and symptom summary given by professional psychiatrists for each dialogue.Finally, we finetune on state-of-art pre-training models and respectively give our dataset baselines on response generation, topic prediction, dialog summary, and severity classification of depression and suicide risk.