AI Model Can Predict Depression Severity From White People’s Facebook Posts
A recent study published in the Proceedings of the National Academy of Sciences revealed race-based differences in the expression of depression in social media language. The research conducted by Sunny Rai, Ph.D., from the University of Pennsylvania, focused on how race moderates the relationship between language features from Facebook posts and self-reported depression.
The analysis included 868 Black and White English-speaking individuals, highlighting that depression severity predicts the usage of first-person pronouns in White individuals but not in Black individuals. Additionally, the study pointed out that more belongingness and self-deprecation-related negative emotions were observed in White participants’ posts.
Despite being trained on similar amounts of data, machine learning models performed less effectively in predicting depression severity among Black individuals compared to White individuals. The findings remained consistent even when the models were exclusively trained on the language of Black individuals.
While emphasizing that social media language and AI models cannot diagnose mental health disorders or replace professional help, Rai stated, They do show immense promise to aid in screening and informing personalized interventions. The need for diverse and representative data was highlighted as crucial for further improvements in integrating AI into research or clinical practice.
In conclusion, the study sheds light on the importance of understanding race-based differences in expressing depression through social media language and the potential of AI models to assist in screening for mental health conditions. Researchers emphasize the necessity of considering diverse datasets to enhance the accuracy and effectiveness of such models in the future.
