Bidirectional LSTM and Attention for Depression Detection on Clinical Interview Transcripts

Existing studies using interviews focus on the automated depression detection task. However, it remains challenging to diagnose depression- requiring time-intensive interviews. Therefore, automated methods could help psychiatric professionals make faster, more informed diagnostic decisions by lingui...

Full description

Saved in:
Bibliographic Details
Published in2022 IEEE 10th International Conference on Information, Communication and Networks (ICICN) pp. 638 - 643
Main Authors Li, Mingzheng, Xu, Haojie, Liu, Weifeng, Liu, Jiangwei
Format Conference Proceeding
LanguageEnglish
Published IEEE 23.08.2022
Subjects
Behavioral sciences
bert
computational paralinguistics
Depression
Interviews
Linguistics
Neural networks
recurrent neural networks
Task analysis
Text recognition
Online AccessGet full text

Cover

More Information
Summary:Existing studies using interviews focus on the automated depression detection task. However, it remains challenging to diagnose depression- requiring time-intensive interviews. Therefore, automated methods could help psychiatric professionals make faster, more informed diagnostic decisions by linguistic patterns in these interviews. We propose a novel method that analyzes interview transcripts to detect depression. We employ bidirectional LSTM to capture contextual information and self-attention to obtain the important information for the participants. Then a pooling layer is introduced to obtain the subject's state representation. Experimental results on the DAIC-WOZ dataset show the F1 performance improvement of our model compared with other methods.
DOI:10.1109/ICICN56848.2022.10006532