Lightme: analysing language in internet support groups for mental health

Area of Science:

• Computational linguistics
• Artificial intelligence in mental health

Background:

• Online support groups require moderation to ensure user safety.
• Automated text classification offers a promising solution for triaging harmful content.

Purpose of the Study:

• To develop and evaluate an automated triage classifier for harmful posts in online mental health forums.
• To identify linguistic features indicative of crisis posts.

Main Methods:

• Utilized Natural Language Processing (NLP) and Machine Learning (ML) techniques.
• Trained a triage post classifier on a dataset from a youth mental health forum (Reachout.com).

Main Results:

• The classifier achieved 52% performance for crisis posts, outperforming state-of-the-art methods.
• Identified six key linguistic characteristics of crisis posts: hopelessness, concise negative emotions, varied emotions, dissatisfaction with services, storytelling, and peer advice seeking.

Conclusions:

• Textual content alone is sufficient to build a competitive triage classifier.
• Further research can enhance classifier performance by translating qualitative and quantitative findings into features.