DepressNet: A Multimodal Hierarchical Attention Mechanism approach for Depression Detection

Abstract

Depression has been the prime cause of mental-health illness globally. A major depressive disorder is a common mental health disorder that affects both psychologically and physically, which could lead to the loss of life in extreme cases. Detection of depression from the recording of an interview could help with early diagnosis. This paper proposes a three-stage framework multimodal machine learning approach called DepressNet for depression detection using the PHQ-8 Questionnaire score. Bidirectional Long Short-Term Memory (BLSTM) layer network has been proposed, and Extended Distress Analysis Interview Corpus (E-DAIC) dataset was used for the training and validation of the proposed method with the uses multiscale temporal features from audio, video, and text modality and attention mechanisms for fusion. The method achieved the RMSE of 4.32 and CCC of 0.662 on the development set, and on the test set, we got RMSE of 5.36 and CCC of 0.457, outperforming the other methods.