A team of eight experts, including Stony Brook University Computer Science Professor, Andrew Schwartz, developed an algorithm that might help predict depression using Facebook posts. The algorithm collects and analyzes the words in a person’s Facebook status to detect symptoms of depression.
“For the first time, a large portion of people are digitally documenting a large portion of their life, and that can be used to help people,” Schwartz said.
According to Statista.com, there were 204 million Facebook users in the United States as of October 2018.
The team analyzed six months of Facebook posts from 683 different patients who visited a large urban emergency room. One hundred and fourteen of those patients already had a diagnosis of depression in their medical records.
Once all of the information was collected, the algorithm counted how frequently each patient used terms associated with depression. Words that denote depressed mood, hostility, loneliness, pain felt in any part of the body or medical references were used to help the algorithm make predictions. Some of the most common things the researchers came across were words like crying, tears, pain, abbreviations like WTF and even the “less than three” emoticon representing a heart.
The researchers found that patients with a diagnosis of depression were more likely to type more words on Facebook. They would usually post an average of 10,655 words over the six months — 3,794 more words than patients without depression. Of that sample, 76.7 percent of individuals were female and 70.1 percent were African Americans.
The goal of the research is to give doctors and hospitals an additional tool for the screening of depression process. According to the resulting research paper, primary care providers have as little as 15 minutes to “address many facets of health within a clinical visit.”
“The assessment, once [patients] come in, is multifaceted,” Susan Morin, director of Adolescent and Eating Disorder Partial Hospitalization at Mather Hospital, said. “The psychiatrists sees them, the nurse practitioner sees them, the social worker sees them. There’s more than one assessment being done.”
Identifying individuals who may develop depression through social media can be a part of this process and might save time.
“It will be a click of a button,” Schwartz said.
But Schwartz also points out that the algorithm is not intended to be used as a substitute for psychiatric evaluation.
Although a sizeable portion of the population had already been diagnosed with depression at some point, Schwartz said he believes the same methodology can be used to create a better process that is representative of the general population.
“Generally, predictive models like this work best for the population that they are train [for], Schwartz said via email. “One would need to train the model for whatever population they planned to apply it, but there’s no reason to think the same process won’t work for others.”