JMIR Publications

ABSTRACT

Background:

Narcissistic Personality Disorder (NPD) affects up to 6.2% of people in the United States. NPD is defined in the DSM-V as a personality disorder marked by grandiose self-importance, a preoccupation with fantasies of unlimited success or power, an excessive need for admiration, a sense of entitlement, exploitative behavior in relationships, and a lack of empathy in relationships. NPD significantly disturbs personal relationships and society through manipulation, abuse and lack of reciprocity. In the digital world, understanding NPD's manifestation in online interactions is vital for mental health professionals to observe the prevalence of NPD and intervene in important cases effectively.

Objective:

This study aims to develop a classification model that can accurately detect narcissistic traits within Reddit posts using text mining methods on post titles and bodies. We focused on detecting posts where narcissistic traits were exhibited or posts with a potentially narcissistic individual mentioned in them.

Methods:

Reddit Posts on Narcissism subreddit were extracted and 966 were manually annotated. As the control group, posts from DeepThoughts, Showerthoughts, and ImposterSyndrome subreddits were downloaded. Features were extracted using term frequency-inverse document frequency (tf-idf), sentiment analysis on post titles and bodies. Alternatively several text embedding models were used. Classification algorithms were applied on the resulting corpus, and the prediction performance was evaluated.
We executed several experiments with different data compositions.

Results:

Our models detected posts exhibiting narcissistic traits with high accuracy and an average F1 score of 0.909 using text-embedding-3-large embedding model. Embedding models consistently outperformed tf-idf and sentiment analysis approaches.

Conclusions:

Our findings show it is possible to detect narcissistic tendencies in Reddit posts. The study tested the prediction model on different subreddits and combinations of data. Training narcissistic-positive posts against regular thoughtful subreddits, it is found to be possible to discriminate narcissistic traits.