Whether you realize it or not, you leave a trail of clues about your mental health on social media. Emerging research suggests that words, characters, and even emoji can reveal information about people’s moods and mental well-being.
That might seem obvious when it comes to explicitly emotional posts, but information about mental health can be embedded in even the most standard of social media messages — and a new field of data science is trying to understand how to detect and interpret those signals.
In a new study published in EPJ Data Science, two researchers accessed the Instagram accounts of 166 volunteers and then applied machine learning to their collective 43,950 images in order to identify and predict depression. By comparing those predictions to each individual’s clinical diagnosis, the researchers discovered that their model outperformed the average rate of physicians accurately diagnosing depression in patients.
“Doctors don’t have visibility into our lives the way our mobile phone does.”
In other words, an Instagram account may have the potential to reveal whether you’re experiencing depression. The right algorithm might just be better at making that prediction than a trained physician.
“Doctors don’t have visibility into our lives the way our mobile phone does,” said Chris Danforth, co-author of the study and Flint Professor of Mathematical, Natural, and Technical Sciences at the University of Vermont. “It knows a lot more about us than we know about ourselves.”
There are, however, a few caveats to the study’s fascinating discovery.
The researchers recruited the volunteers through Amazon Mechanical Turk, an online marketplace that matches workers with businesses and developers. Once the volunteers learned that the study required letting the researchers access their Instagram accounts, more than half of them dropped out. The small sample size and lack of demographic information mean it’s impossible to generalize the findings to the larger population.
The study analyzed color, filters, face detection, and user comments and engagement to make its predictions.
Photos posted by depressed users were more often bluer, darker, and grayer, hues that previous research has associated with negative mood. They also were less likely to use Instagram filters, but disproportionately chose the black-and-white “Inkwell” filter when they did take advantage of the tool. In contrast, “healthy” participants favored the tint-lightening “Valencia” filter. Meanwhile, the more comments a post received, the more likely someone with depression had posted it.
The researchers also enlisted a second group of volunteers to rate the photos for “happiness, sadness, likability, [and] interestingness.” The human assessments for sadness and happiness predicted which participants experienced depression, but they didn’t correlate with the mental health signals picked up by machine learning.
That’s a promising finding because it means algorithms may excel at detecting conditions where humans fail.
“There has been anecdotal and preliminary evidence that these sorts of signals are present and relevant to mental health, but this study provides compelling evidence of their utility,” said Glen Coppersmith, founder and CEO of the startup mental health analytics company Qntfy. (Coppersmith wasn’t involved in the study.)
“It’s important that computers can do this.”
Danforth said he can imagine a future in which people opt to download an app that analyzes their social media for signs of emotional or psychological distress and sends a message to an individual’s doctor when they need to be seen by a mental health professional.
That future, of course, requires a lot more research — and the trust of social media users. Allowing an app to collect and analyze social media data in the context of mental health raises serious questions about about security. In particular, that information could never get into the hands of insurers and potential employers, who might use it to make decisions about a person’s health care premiums or employment.
But Danforth believes such technology serves an important public good.
“It’s important that computers can do this,” he said. “It’s better if we can get somebody who [might] die by suicide in 2018 in front of a psychologist sooner because there’s something about their social media that made it clear to the machine that they needed help and it wasn’t obvious to the people around them.”
That might not be a possibility you imagined when you first started uploading pictures to Instagram, but, if successful, an algorithm like that could change the way we detect and treat mental health conditions in the 21st century.
If you want to talk to someone or are experiencing suicidal thoughts, text the Crisis Text Line at 741-741 or call the National Suicide Prevention Lifeline at 1-800-273-8255. Here is a list of international resources