Department of Computer Engineering
Computational Assessment of Depression Severity using Behavioral Cues
Ergun Batuhan Kaynak
(Supervisor: Asst. Prof. Dr. Hamdi Dibeklioğlu)
Computer Engineering Department
Major Depressive Disorder (MDD) is a mental illness which affects the mental and physical functioning of the individual. With the increase in suicide rates linked to depression, techniques for its detection and cure became a prominent topic. Due to the lack of data and the nature of the existing data, detecting the severity of depression is still an open research area. We propose a multimodal depression severity prediction model that predicts the depression severity of a participant according to the PHQ8 scale. We use the Audio/Visual Emotion Challenge (AVEC 2019) challenge dataset, E-DAIC, which consists of video, audio and text features of clinical interviews for psychological distress conditions such as anxiety and depression. We will use recurrent neural networks to learn important depression cues from video and audio, combine learning strategies to further differentiate cues within their context and experiment with different fusion techniques to best combine our features on different levels. Although quite indicative of depression severity, text modality is only explored primitively in the current literature. To explore this area, we will work on extracting summaries, sentence encodings, sentiment analysis and many more features from text.
DATE: 04 November 2020, Wednesday @ 13:35