PROVO — Twitter could be a canary in the mine, providing early warning that someone is at risk for suicide.
That's according to BYU research that used computer technology to sift through millions of tweets looking for words that hint at suicide risk, then compared the results with real suicide data in each state. The suicide rates and the number of troubling tweets were strongly correlated, according to Christophe Giraud-Carrier, associate professor of computer science at Brigham Young University, who is one of the study's seven authors.
The findings are being published in the journal Crisis on Friday.
"This is an innovative and novel way to follow a public health problem," said Carl Hanson, associate professor and director of the master of public health program at BYU.
Researchers with backgrounds in health and computer science teamed up to identify and then search for key words that define suicide risk factors. Over three months, the computer scientists screened millions of tweets and identified 37,717 that were "genuinely troubling" from 28,088 unique users about whom they had some geographically identifying information, the scientists explained in background to the study. Then they looked at each state's ratio of suicidal tweets compared to actual suicide rates and found they were "strongly correlated."
"We see it's possible to monitor risky conversations occurring through Twitter, and they tend to mirror what we see in CDC data on actual suicides," said Hanson.
Increasingly, social and other media are playing a role in epidemiology, he said. This kind of research even has a name now: "infomediology." He predicts it will prove most useful when combined with other sources to look at risk and possibility. Once risk is identified, the question becomes whether it's possible to intervene.
Besides using certain search terms for inclusion criteria, they also used some terms as exclusion criteria. For instance, if a tweet about "I hate life" also had "failed my test," it wasn't considered a genuine suicidal thought, said Giraud-Carrier. "That person is not going to kill himself. He's mad because of a poor grade." Through a process of identifying terms that negate threat, "we hope we have gotten rid of enough noise to make things more legitimate."
The group of scientists has been collaborating with a Utah suicide prevention coalition and believes that "what we see online is correlated with what happens in real life," he said, adding it's possibly particularly applicable to teenagers.
The study found states like Alaska or Utah, which have higher-than-average suicide rates, also had more Twitter users who posted tweets that might indicate risk. In Utah, for instance, researchers found 195 Twitter users who might be at risk, while they saw more than 3,000 in Texas, which has a suicide rate that's a bit lower but a much larger population.
"Tweets may be useful to address some of the functions that suicide hotline groups perform, but at the discretion and potential for such organizations to provide those services via Twitter," said Michael Barnes, health science professor and study co-author, in a written statement about the research.
According to background material provided with the study, about 15 percent of tweets give at least some state-level location information, which allowed the comparison of risky sounding tweets to the actual suicide rates.
"The hope is that this will expand what's being done by psychologists, schools and others working in prevention," said Giraud-Carrier.
Other researchers included BYU health science professor Josh West and students Jared Jashinksy, Scott Burton and Trenton Argyle.
EMAIL: [email protected], Twitter: Loisco