Today’s teens are always on their phones, a transition that has resulted in teens discussing suicidal risk, psychological distress and sleep disturbances online. Often, teens feel more comfortable discussing difficult topics such as these online rather than in person.
With this background in mind, SPARC-Life’s Drs. Jamie Zelazny and Candice Biernesser are collaborating with Georgia Tech’s Dr. Munmun De Choudhury to apply a machine-learning algorithm to teens’ social media use. Dr. De Choudhury has developed algorithms that can identify risk for suicide from adolescent's social media data. The algorithm picks up on linguistic changes such as self-focus, reduced social engagement, poor language ability, and expressions of certain negative moods as red flags for potential suicidal ideation. The algorithm currently sorts information from Twitter, Facebook, Instagram, Reddit, and Google search history.
Drs. Zelazny and Biernesser will be using this algorithm in their own project at the University of Pittsburgh with collaborators at Children's Hospital of Philadelphia. With funds from the University of Pittsburgh's ETUDES Center, an ALACRITY Center funded by NIMH, they will investigate the feasibility and effectiveness of analyzing this type of data for risk factors for suicidality. The study will recruit 500 youth who are at high risk for suicide from another ETUDES project and ask them to share their social media data and Google search history from the past year. The teens who enroll will download their data and upload it to Dr. De Choudhury’s secure website, stripped of identifying information.
Then, the research team at Georgia Tech will analyze the data and extract information on mood, emotional impulsivity, online victimization, social involvement, feelings of hopelessness as reflected in verb tense, interpersonal awareness as reflected in pronoun use, and sleep patterns as reflected in time spent online. Dr. De Choudhury’s team will then debut new factors that may be associated with suicidal ideation such as verbal coherence, term-frequency, information seeking, and shifts in discussion topics. The overall goal is to investigate if social media data patterns will emerge in predicting suicidality.