Detecting Narcissistic Personality Disorder Traits on Forums: Proof-of-Concept Study

Area of Science:

• Computational linguistics
• Psychiatry
• Machine learning

Background:

• Identifying narcissistic personality disorder (NPD) traits is challenging, necessitating novel detection methods.
• Online forums offer rich data for understanding mental health, but require advanced analytical tools.
• Early detection of NPD traits can significantly improve patient outcomes.

Purpose of the Study:

• To develop and evaluate a machine learning model for detecting language patterns associated with NPD traits in Reddit posts.
• To identify both authors exhibiting NPD traits and posts discussing individuals with these traits.
• To assess the reliability of automated detection of NPD-related language.

Main Methods:

• Analysis of 75,985 posts from four distinct Reddit communities.
• Annotation of 966 posts by two psychiatrists to create a reliable dataset.
• Training machine learning models, including transformer-based embedding models, to differentiate NPD-trait-related posts from general content.

Main Results:

• Machine learning models achieved high accuracy, with modern embedding models reaching a mean F1-score of 0.90.
• Models effectively identified NPD-trait markers, balancing recall and precision while minimizing false positives.
• Performance remained robust even after removing common keywords like "narcissism".

Conclusions:

• Automated analysis of online posts presents a viable approach for identifying NPD traits.
• This technology holds potential for future clinical research applications.
• Strict ethical oversight is crucial, and the tool is intended for research, not public diagnosis.