[4Xin1] Poster session 2

[4Xin1-50] utomatic Depression Severity Classification and Feature Importance Analysis Using Mental Disease Dialogue Corpus

〇Sho Katsuki¹, Hiroya Tanaka⁴, Hironobu Nakamura², Taishiro Kishimoto³, Yoshinobu Kano¹ (1.Shizuoka Univercity, 2.Tokyo Medical and Dental University, 3.Keio University, 4. Natural Institute of Technology Kurume College)

Keywords:Mental Health, Disease Type Prediction

We have been building our UNDERPIN mental disease dialogue corpus, which includes disease types, disease severity tests, and dialogue voices with their transcriptions and audio/linguistic annotations. We trained our classifier to classify depression patients using this corpus.
We defined four classes (healthy, light level, more than middle level, and recovered), then tried six pairs of binary classification of these four classes. Our classification results show that we can classify light level patients with more than middle level patients. We can also classify recovered people with healthy people. We confirmed new features that contributed to the classifications. In future, we plan to add new features, and create automatic annotation tools.