AI analysis of social media language predicts depression severity among white Americans, but not among black Americans

When the researchers used a standard language-based computer model to analyze Facebook posts, they were able to predict the severity of depression in white people, but not in black people. Words and phrases associated with depression, such as first-person pronouns and negative emotion words, predicted depression severity about three times more for whites than for blacks.

This research today Proceedings of the National Academy of Sciencesis co-authored by researchers at the University of Pennsylvania in Philadelphia and the National Institute on Drug Abuse (NIDA), part of the National Institutes of Health (NIH).

Previous research has shown that the language of social media can provide useful information as part of mental health assessments, but the results of this study show that the language used by depressed patients By highlighting key demographic differences, we point out potential limitations in generalizing the language of social media. The results also highlight the importance of including diverse data pools to ensure accuracy in the development of machine learning models, which are applications of artificial intelligence (AI) language models.

“As society seeks to leverage AI and other technologies to provide much-needed mental health care, no one is left out or misrepresented,” said NIDA Director Nora Volkow, MD. We have to make sure that doesn’t happen.” She said, “More diverse datasets are essential to ensure that health care disparities are not perpetuated by AI, and these new technologies can help customize more effective medical interventions.”

The study recruited 868 consenting participants who identified themselves as black or white, and found that a model trained on Facebook language used by white participants with self-reported depression was When tested on people, it was demonstrated that it showed strong predictive performance. However, when the same model was trained on her Facebook language from Black participants, it performed poorly when tested on Black participants and only marginally better when tested on White participants.

Among White participants, depression severity was associated with increased use of first-person singular pronouns (“me,” “me,” “me”), but this correlation was not found among Black participants. I couldn’t see it. Additionally, white people have a strong sense of belonging (“weird,” “creepy”), self-criticism (“messed up,” “terrible”), insecurity and being an outsider (“scared,” “misunderstood,” etc.). self-deprecation (“worthless,” “worthless”) and hopelessness (“begging,” “hollow”) increased with increasing severity of depression, but no such correlation was observed among blacks. There was no relationship. Clinicians have noticed demographic differences in how people express symptoms of depression for decades, but this study shows how this plays out on social media. I am.

Language-based models show promise as personalized, scalable, and affordable tools for screening mental health disorders. For example, excessive self-referential language, such as the use of first-person pronouns, and negative emotions, such as self-deprecating language, are often considered clinical indicators of depression.

However, racial and ethnic considerations are severely lacking when assessing mental disorders through language, leading to inaccurate computer models. Despite evidence showing that demographic factors influence the language people use, previous research has not investigated how race and ethnicity affect the relationship between depression and verbal expression. We did not systematically investigate whether to give.