Using GPT-4 to Identify Social Anxiety from Interviews
Author Information
Author(s): Julia Ohse, Bakir Hadžić, Parvez Mohammed, Nicolina Peperkorn, Janosch Fox, Joshua Krutzki, Alexander Lyko, Mingyu Fan, Xiaohu Zheng, Matthias Rätsch, Youssef Shiban
Primary Institution: PFH University of Applied Sciences, Göttingen, Germany
Hypothesis
Can GPT-4 accurately infer social anxiety disorder (SAD) symptom severity from semi-structured interview data?
Conclusion
GPT-4 can effectively predict social anxiety symptoms from interview data, showing a high correlation with self-reported measures.
Supporting Evidence
- GPT-4 predictions were highly correlated with self-reported social anxiety scores (r=0.79).
- An F1 accuracy score of 0.84 was achieved in predicting social anxiety.
- Only 35% of individuals with social anxiety receive a diagnosis, highlighting the need for better detection methods.
Takeaway
Researchers used a computer program called GPT-4 to understand how anxious people feel in social situations by talking to them online, and it worked really well!
Methodology
Participants were interviewed online, and their responses were analyzed by GPT-4 to predict their social anxiety scores.
Potential Biases
The sample was predominantly from a single cultural background, which may limit the generalizability of the results.
Limitations
The study relied on self-reported measures rather than professional diagnoses, which may affect the validity of the findings.
Participant Demographics
Participants included 23 males and 20 females, aged 20 to 69, with a mean age of 31.19 years.
Statistical Information
P-Value
0.001
Statistical Significance
p<0.001
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website