Models of Gender Dysphoria Using Social Media Data for Use in Technology-Delivered Interventions: Machine Learning...

Area of Science:

• Clinical Psychology
• Computational Linguistics
• Digital Health

Background:

• Gender dysphoria (GD) optimal treatment involves medical intervention, yet transgender and nonbinary individuals encounter significant barriers accessing care.
• Untreated GD correlates with adverse mental health outcomes, including depression, anxiety, suicidality, and substance misuse.
• Technology-delivered interventions offer discrete, safe, and flexible access to psychological support for GD distress.

Purpose of the Study:

• To evaluate the preliminary effectiveness of machine learning (ML) and natural language processing (NLP) in modeling gender dysphoria.
• To utilize social media data from transgender and nonbinary individuals for modeling gender dysphoria.

Main Methods:

• Employed 6 ML models and 949 NLP variables to analyze 1573 Reddit posts from transgender and nonbinary forums.
• Utilized qualitative content analysis with a clinician-informed codebook to identify gender dysphoria as the dependent variable.
• Transformed linguistic content into predictors using NLP techniques (n-grams, LIWC, word embeddings, sentiment, transfer learning) for ML algorithms.

Main Results:

• An optimized extreme gradient boosting (XGBoost) ML model achieved high accuracy (0.84), precision (0.83), and speed (1.23 seconds) in modeling gender dysphoria.
• Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5) clinical keywords were the most predictive NLP-generated variables.
• Common misclassifications involved posts expressing uncertainty, unrelated stressors, identity exploration, or discussing body image.

Conclusions:

• ML and NLP models demonstrate significant potential for integration into technology-delivered interventions for gender dysphoria.
• Findings support the value of incorporating ML and NLP in clinical science, particularly for studying marginalized populations.