GPT-4 shows potential for identifying social anxiety from clinical interview data

Area of Science:

• Computational linguistics
• Psychiatry
• Artificial Intelligence in mental health

Background:

• Extensive research exists on AI and Natural Language Processing (NLP) for depression symptom detection.
• Limited studies have explored AI's potential for identifying social anxiety symptoms.
• Social anxiety disorder (SAD) significantly impacts individuals' quality of life.

Purpose of the Study:

• To evaluate the efficacy of the large language model (LLM) GPT-4 in inferring social anxiety symptom severity.
• To assess GPT-4's performance using transcripts from semi-structured interviews.
• To determine the correlation between GPT-4's predictions and self-reported social anxiety measures.

Main Methods:

• Recruited 51 adult participants from a German convenience sample.
• Participants completed a social anxiety self-report questionnaire (SPIN).
• Interview transcripts were analyzed by GPT-4 to predict social anxiety symptom strength.

Main Results:

• GPT-4's predictions demonstrated a high correlation (r=0.79) with participants' self-reported social anxiety scores.
• An F1 accuracy score of 0.84 was achieved, adhering to established cut-off conventions.
• The study highlights the strong potential of LLMs in quantifying social anxiety from textual data.

Conclusions:

• GPT-4 shows significant capability in assessing social anxiety symptom severity from interview transcripts.
• This AI-driven approach offers a reliable and accurate method for social anxiety assessment.
• Future research should validate these findings in larger, more diverse populations.