A New Way To Listen

New research published in the journal Suicide and Life-Threatening Behavior shows how machine learning can help identify suicidal behavior using a person’s spoken or written words. The technology was able to pinpoint which participants in the study were suicidal, mentally ill but not suicidal, or neither in the vast majority of cases.

John Pestian and a team of researchers studied 379 patients from emergency departments and inpatient and outpatient centers at three locations between Oct. 2013 and March 2015. The patients, who were classified as suicidal, mentally ill but not suicidal, or neither (serving as the control group), answered standardized behavioral rating tests and took part in a semi-structured interview in which they were asked five open-ended questions such as “Do you have hope?” and “Are you angry?” to stimulate conversation.

The researchers then pulled verbal and non-verbal language (e.g., laughs, sighs, etc.) from the gathered data and used machine learning algorithms to analyze it. The algorithms correctly identified suicidal persons with 93 percent accuracy and were 85 percent accurate in pinpointing a person who was suicidal, had a mental illness but was not suicidal, or neither.