Depression in Black people goes unnoticed by AI models analyzing language in social media posts

Researchers at the University of Pennsylvania have discovered that methods used to detect depression through social media posts may not be effective when applied to posts by Black individuals. The study, published in the journal PNAS, highlights the importance of considering the intersection of race, health risks, and social media when developing mental health interventions.

Previous research had found that the use of first-person pronouns (“I”) and certain categories of words, such as self-deprecating terms and expressions of feeling like an outsider, in social media posts could predict depression among users. However, when analyzing Facebook posts from over 800 people, with equal numbers of Black and white individuals, some with reported depression and some without, the researchers found that these “predictive” words mainly applied to white social media users.

“We were surprised that these language associations found in numerous prior studies didn’t apply across the board,” said Sharath Chandra Guntuku, PhD, a researcher in the Center for Insights to Outcomes at Penn Medicine and an assistant professor (research) of Computer and Information Science in Penn Engineering. “We need to have the understanding that, when thinking about mental health and devising interventions for treatment, we should account for the differences among racial groups and how they may talk about depression. We cannot put everyone in the same bucket.”

The researchers found that when the previously identified predictive words were used in an artificial intelligence (AI) model, it performed strongly among white individuals but was more than three times less predictive for depression when applied to Black Facebook users. Even when the AI model was trained on language used by Black individuals in their posts, it still performed poorly.

“Why? There could be multiple reasons,” said Sunny Rai, PhD, the study’s lead author and a postdoctoral researcher in Computer and Information Science. “It could be the case that we need more data to learn depression patterns in Black individuals compared to white individuals. It could also be the case that Black individuals do not exhibit markers of depression on social media platforms due to perceived stigma.”

The researchers also found that Black individuals tended to use “I” more frequently in their posts, including those who did not report having depression. Additionally, while self-deprecating and outsider-related words were associated with depression in white individuals, these language groups were not specifically tied to depression in Black individuals.

Rai emphasized the need for increased representation of Black individuals and other races and ethnicities in studies to better understand how depression is expressed across different groups. This understanding could lead to the development of better predictive models and improved mental health interventions.

“AI-guided models that were developed using social media data can help in monitoring the prevalence of mental health disorders, especially depression, and their manifestations,” Rai said. “Such computational models hold promise in assisting policy-making as well as designing AI assistants that can provide affordable yet personalized healthcare options to citizens.”

Lyle Ungar, PhD, a co-author on the study and professor of Computer and Information Science, added, “Understanding differences in how Black and white people with depression talk about themselves and their condition will be important when training psychotherapists who work across different communities.”

The researchers plan to expand their study to examine how depression is expressed in cultures beyond the United States, furthering their understanding of the differences in how mental health conditions are outwardly expressed.

See original

Depression in Black people goes unnoticed by AI models analyzing language in social media posts

Methods researchers developed to detect possible depression through language in social media posts don’t appear to work when applied to posts by Black people on social media, according to a new analysis by researchers from Penn’s Perelman School of Medicine and its School of Engineering and Applied Science. The research, published in PNAS, points to an area to focus on for significant improvement and amplifies the importance of considering the intersection of race, health risks, and social media.

Work in the past uncovered that using first-person pronouns in posts (“I”) and certain categories of words (self-deprecating terms and expressing outsider feelings) in social media posts was predictive of depression among people who use social media. However, in analyzing Facebook posts from more than 800 people—a sample that included equal numbers of Black and white individuals, some who reported having depression and some who did not—the researchers found that the predictive qualities of the “predictive” words applied mainly to white people      on social media.

“We were surprised that these language associations found in numerous prior studies didn’t apply across the board,” said one of the study’s senior authors, Sharath Chandra Guntuku, PhD, a researcher in the Center for Insights to Outcomes at Penn Medicine and an assistant professor (research) of Computer and Information Science in Penn Engineering. “We need to have the understanding that, when thinking about mental health and devising interventions for treatment, we should account for the differences among racial groups and how they may talk about depression. We cannot put everyone in the same bucket.”

When the types of words identified in the past as predictive for depression were plugged into an artificial intelligence-guided model, the researchers found that it performed “strong[ly]” among white people. However, they found that the model was more than three times less predictive for depression when applied to Black people who use Facebook.

Even when the researchers trained the artificial intelligence (AI) model on language used by Black people in their posts, the model still performed poorly.

“Why? There could be multiple reasons,” said the study’s lead author, Sunny Rai, PhD, a postdoctoral researcher in Computer and Information Science. “It could be the case that we need more data to learn depression patterns in Black individuals compared to white individuals. It could also be the case that Black individuals do not exhibit markers of depression on social media platforms due to perceived stigma.”

Something that potentially confounded the existing depression-detection models, the researchers found, was that Black people tended to use “I” more overall in their posts. That includes participants in the study who did not report having depression.

And while words expressing self-deprecation (including “worthless” and “useless”) and feeling like an outsider (such as “weirdo” and “creep”) were associated with white people with depression, these groups of language were not specifically tied to depression in Black people.

Rai said that it’s clear that researchers need to increase representation of Black people and other races and ethnicities in studies to better understand the ways in which depression is expressed across different groups of people. Through this, the researchers hope that better predictive models can be established and mental health needs can be better addressed.

“AI-guided models that were developed using social media data can help in monitoring the prevalence of mental health disorders, especially depression, and their manifestations,” Rai said. “Such computational models hold promise in assisting policy-making as well as designing AI assistants that can provide affordable yet personalized healthcare options to citizens.”

Insights made through AI can also serve the education of professionals who help people manage depression.

“Understanding differences in how Black and white people with depression talk about themselves and their condition will be important when training psychotherapists who work across different communities,” said, Lyle Ungar, PhD, a co-author on the study and professor of Computer and Information Science.

In the vein of better understanding the differences in how mental health conditions are outwardly expressed, Guntuku, Rai, and Ungar are planning to study how depression is expressed in cultures beyond the United States.

The PNAS study was funded, in part, by the National Institute of Drug Abuse (ZIA-DA000628), the National Institute on Minority Health and Health Disparities (R01MD018340) and the National Institute on Alcohol Abuse and Alcoholism (R01 AA028032-01). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.