Are Your Texts Depressed? The Computer Knows, Maybe
Chronicle of Higher Education, Wired Campus Blog: June 18, 2010 - Matthew Kalman
Software may know when you are depressed by examining your online behavior. Researchers at Ben-Gurion University of the Negev, in Beer-Sheva, Israel, have developed a program that can detect depression in online texts and could serve as a screening tool to direct potential patients towards treatment. Psychologists caution, however, that it hasn't actually been tested on real people.
Yair Neuman, associate professor in the department of education at Ben-Gurion, led a team that developed a computer program capable of identifying language with signs of depression. In a test, the program was used to scan more than 300,000 English-language texts from blogs and from online queries that people posted to mental-health Web sites. After the program identified the texts as depressive, a panel of four clinical psychologists reviewed 200 examples of such writings. There was a 78-percent correlation between the verdict of the computer program and the analysis of the human panel.
Prof. Neuman said the program was designed to find depressive content hidden in language that did not mention obvious terms like “depression” or “suicide.” He suggested that the program could be used to carry out initial screening on texts written by people who didn’t even realize they might have a problem.
“The problem is that most people are not aware of their situation and they will never get to an expert psychologist,” says Prof. Neuman. “The system can provide a screening process that will raise the awareness of the depressed and will send them to an expert because we cannot actually replace excellent human judgment. What we can do is to provide a very efficient tool for screening for depression.
“In the United States, for instance, there is a huge problem of people suffering from depression and they are not diagnosed. The usual screening procedure is a questionnaire you should fill in online, but it is a self-selective process," he says, noting that people who fill out such a survey already suspect they have a problem.
"What we can do is to analyze proactively, and this is the difference." Web sites focusing on consumer mental health might install the tool, and users could see a pop-up warning if the comments they post indicate a depressive pattern.
But those warnings might be false, says one mental-health professional. “Psychiatric diagnosis is a very, very complicated issue,” warns Tuvia Peri, director of the community counseling clinic in the department of psychology at Bar-Ilan University. “You don’t have a very high level of agreement between professionals because the diagnosis of depression is quite vague.”
“There is a long history of trying to determine psychiatric diagnosis by computers, to try to make them efficient and fast," he says. "This is a very small step forward, an important step, but we have texts that were diagnosed by a machine and then we have the same texts diagnosed by clinical psychologists. We don’t have any data about the actual, real state of the people who have written these texts.”