Detecting Psychological Distress in Adults Through Transcriptions of Clinical Interviews

Automatic detection of psychological distress, namely posttraumatic stress disorder (PTSD), depression, and anxiety, is a valuable tool to decrease time, and budget constraints of medical diagnosis. In this work, we propose two supervised approaches, using global vectors (GloVe) for word representation, to detect the presence of psychological distress in adults, based on the analysis of transcriptions of psychological interviews conducted by a health care specialist. Each approach is meant to be used in a specific scenario: online, in which the analysis is performed on a per-turn basis and the feedback from the system can be provided nearly live; and o✏ine, in which the whole interview is analysed at once and the feedback from the system is provided after the end of the interview. The online system achieves a performance of 66.7% accuracy in the best case, while the o✏ine system achieves a performance of 100% accuracy in detecting the three types of distress. Furthermore, we re-evaluate the performance of the o✏ine system using corrupted transcriptions, and confirm its robustness by observing a minimal degradation of the performance.