Mining social media to learn how race influences depression descriptions

2024

Challenge

Depression is a leading mental health concern worldwide and its symptoms are often reflected in patients’ language. Social media offers a scalable platform for identifying depression with the help of artificial intelligence (AI), but previous research has largely overlooked racial differences in how depression is expressed linguistically, focusing instead on other demographic distinctions like age and sex. AI tools that do not account for racial variations in language are limited in their ability to identify signs of depression in certain groups, thereby exacerbating health disparities.

Advance

A team of IRP researchers and their collaborators revealed differences between how depression affects the language Black and white individuals use on social media. White participants with more severe depression tended to use more first-person singular pronouns like “I” and discuss specific topics related to negative emotions, such as self-criticism and feelings of worthlessness. However, Black individuals’ language did not show these same patterns. Moreover, AI algorithms trained via machine learning to predict depression using language data performed well for white individuals but poorly for Black individuals, even when the algorithms were trained exclusively on data from Black participants.

Impact

The findings challenge assumptions about universal linguistic markers of depression and highlight the need for more inclusive AI models that reflect the experiences of diverse populations. By revealing differences in how depression manifests in language, the study calls for more inclusive datasets and culturally sensitive AI models to improve fairness and accuracy in mental health screening. The development of such AI-based tools would represent a critical step toward personalized, inclusive mental health interventions that leave no one behind.

View All Health Topics