Mental Health Severity Detection from Psychological Forum Data using Domain-Specific Unlabelled Data

Area of Science:

• Computational linguistics
• Mental health informatics
• Machine learning for healthcare

Background:

• Mental health assessment is challenging due to privacy and lack of objective measures.
• Subjective patient accounts, often in text, are primary data sources.
• Online platforms and social media offer valuable, albeit unstructured, mental health data.

Purpose of the Study:

• To develop and evaluate an automated system for classifying mental health severity using machine and deep learning.
• To leverage both labeled and unlabeled online text data for improved severity classification.
• To compare the performance of the developed system against existing state-of-the-art methods.

Main Methods:

• Utilized CLPsych 2016 and 2017 datasets from ReachOut online forums.
• Implemented supervised and semi-supervised embedding methods with ReachOut and WebMD corpora.
• Integrated metadata, syntactic, semantic, and embedding features for classification into four severity levels (green, amber, red, crisis).

Main Results:

• The developed automated system achieved high performance on the CLPsych datasets.
• Maximum micro-averaged F-scores of 0.86 for CLPsych 2016 and 0.80 for CLPsych 2017 were obtained.
• The system outperformed other state-of-the-art methods on the ReachOut dataset.

Conclusions:

• Automated text analysis using machine learning is a viable approach for mental health severity assessment.
• The integration of diverse features significantly enhances classification accuracy.
• This methodology offers a scalable solution for analyzing large volumes of online mental health data.