Scientists Create AI Model That Predicts Mental Disorders Based on Reddit Posts

A team of researchers at Dartmouth College has built an artificial intelligence (AI) model that’s able to determine if somebody has a mental disorder based on their Reddit posts.

In a new paper revealed at the 20th annual International Conference on Web Intelligence and Intelligent Agent Technology, a team of researchers from Dartmouth College in New Hampshire describe an AI model able of detecting whether or not somebody suffers from one of three types of mental disorders based solely on their Reddit post history. The researchers say the model—which they actually “trained” using, in large part, posts from Twitter—is able to detect whether or not somebody is suffering from, say, bipolar disorder—with better than 80% accuracy.

As Dartmouth itself notes the model for detecting mental disorders using conversations on Reddit is “part of an emerging wave of screening tools that use computers to analyze social media posts and gain an insight into people’s mental states.” The college reports that “What sets the new model apart is a focus on the emotions rather than the specific content of the social media texts being analyzed.”

The researchers, including Machine Learning PhD Xiaobo Guo et al., open the study by noting that:

“According to the World Health Organization (WHO), one in four people will be affected by mental disorders at some point in their lives. However, in many parts of the world, patients do not actively
seek professional diagnosis because of stigma attached to mental illness, ignorance of mental health and its associated symptoms. In this paper, we propose a model for passively detecting mental
disorders using conversations on Reddit.
Specifically, we focus on a subset of mental disorders that are characterized by distinct emotional patterns (henceforth called emotional disorders): major depressive, anxiety, and bipolar disorders.”

The researchers say they had approximately 2,000 users for each of their three mental disorder “classes,” which included: anxiety disorders (AD), major depressive disorder (MDD), and bipolar disorder (BD). There was also a fourth class: the control group, which contained Reddit users without any identifiable mental disorder. (Yes, apparently they exist.)

Guo et al. identified which users had one of these three disorders based on their own posted diagnosis; that is, the researchers found Reddit and Twitter users who had previously posted phrases such as “I am diagnosed with bipolar [or depression, or anxiety]” and then assigned them to one of the disorder classes. The data from each of the four classes was then split into training the AI model (70%), validation of the AI model (15%), and testing of the AI model (15%).

A look at the Reddit homepage.

To detect one of the three mental disorders in the testing data, the team’s model relied on “unique patterns of emotional transitions”—such as rapid mood swings for bipolar disorder; persistent sad mood for major depressive disorder, and excessive fear and anxiety for anxiety disorders. More specifically, the team’s model looked for unique emotional transitions between posts that were themselves either joyful, angry, sad, etc.

Using these signature “fingerprints” the team’s AI model scanned through the 15% of testing data and was able to identify users with bipolar disorder 85% of the time; users with manic depressive disorder 80% of the time; and users with anxiety disorders 84% of the time.

“Social media offers an easy way to tap into people’s behaviors,” Guo said in Dartmouth’s press release. He added the data is voluntary and public, and published for others to read.

Worth noting—aside from the potential implication of an AI that can determine one’s mental status based on online posts—is the fact the researchers’ results showed that “emotionally healthy users are more likely to present with states that involve anger” than those who are not emotionally healthy. Guo et al. say this is the case because “Research shows that people with discrepant self-esteem are more likely to exhibit greater anger suppression, a more depressive attributional style, and more nervousness… [Thusly] When facing self-esteem threatening stimuli, people with emotional disorders tend to suppress anger and instead transition into sadness.”

In the future, the researchers say they plan to “further validate [their] emotion representations” by “exploring a wider range of classifiers”; that is, a wider range of mental disorders. The researchers also say they want to explore “new methods” of characterizing the emotional states of social media users. Along with more methods, there are aims to add other “modalities” into the predictive mix with images, video, and audio clips.

Feature image: NASA/Scott Kelly

(Visited 7 times, 1 visits today)

Accessibility Toolbar