[Defense] Development of Deep Learning Models for Depression Detection in Texts
Friday, June 9, 2023
11:00 am - 12:00 pm
In
Partial
Fulfillment
of
the
Requirements
for
the
Degree
of
Doctor
of
Philosophy
Steve
Aigbe
will
defend
his
proposal
Development
of
Deep
Learning
Models
for
Depression
Detection
in
Texts
Abstract
Depression is a major mental health disorder affecting a significant portion of the world population. Methods mostly being employed for depression detection are clinical interviews and questionnaire surveys where psychiatric assessment tables are used to establish mental disorder prognosis. However, the shortage of mental health specialists and the limited resources available to primary care physicians who often manage patients with depression prevent close monitoring of symptoms, delaying optimal treatment, and potentially prolonging suffering. These factors indicate a strong need to better measure depression to improve its diagnosis and treatments. Analyzing texts written by an individual can serve as an additional knowledge source to diagnose depression. Consequently, the identification of depressed and non-depressed individuals based on social media posts has become the focus of recent research. Detecting depression in social media and other texts can help in creating early warning systems for depression detection; it can also be used to gather important statistics about depression symptoms and other factors that are associated with depression. Depression detection from social media text data involves information extraction using techniques from natural language processing (NLP) and Artificial Intelligence (AI). For example, Yates et al., in their work on detecting depression and self-harm in the Reddit online forum, proposed a convolutional neural network (CNN)-based model and developed a large-scale depression dataset (RSDD) to evaluate the model. Also, Cohan et al. proposed a system for classifying posts in online mental health forums based on their severity. The authors used a combination of linguistic features and machine learning models to predict the severity of posts. Our research centers on answering the following research questions: (i) What are good architectures for deep learning models to improve the accuracies of depression detection systems in texts? (ii) How can the presence of different depression symptoms be determined, and their severity assessed in texts? (iii) How can we augment (usually very small) depression datasets to alleviate overfitting when using deep learning approaches? In answering question (i), we develop a deep learning model for the detection of depression from social media and online forums textual data. In our preliminary work, we proposed a language model- and CNN-based architecture with the introduction of depression indicators-based weighting scheme and an attention model to develop a depression detection model with performance metrics that rival the reported state-of-the-art results. Our research is mostly based on the use of a BERT language model-based architecture as our previous research on the use of language model-based domain specific wording embeddings for text classification showed improvement in performance metrics. For question (ii), we plan to develop a system based on the presence of depression indicators in depressive texts where we compute a relevance score to indicate the presence of depression indicators that align with the established depression symptoms categories. For question (iii), the plan is to develop a data augmentation (DA) method to create a large dataset from available small-sized depression datasets. The review and comparison of existing DA methods we conducted in our preliminary work shows the need for the development of a generalized DA method. In summary the development of better depression prediction models and the creation of large datasets to train those models are critical for successfully detecting depression and depression symptoms in texts.
Friday,
June
9,
2023
11:00AM
-
12:00PM
CT
Online via
Dr. Christoph F. Eick, proposal advisor
Faculty, students, and the general public are invited.
