It is not primarily a question of whether people tweet about suicide, but how.
A new study by Complexity Science Hub Vienna and MedUni Vienna shows that postings about suicide can actually have a protective effect.
“If the number of tweets with suicide prevention information increases, the number of suicides decreases and more people call help facilities to ask for support,” Hannah Metzler from the CSH explains the results of the new study. “Especially tweets about one’s own coping with a suicidal crisis show enormous potential to encourage people in a similar life situation to get in touch with a help facility,” Thomas Niederkrotenthaler cites a key detail of the research. Such tweets are rarely posted, but whenever they are, a relatively large number of people seek help afterwards.
MEDIA COVERAGE DOES NOT INCREASE SUICIDE RATE
Moreover, the new study again suggests that news about suicide cases does not generally lead to an increase in the suicide rate. “All postings about suicide cases taken together do not show a recognisable increase,” explains Metzler. What matters is how it is reported and about whom. Studies have shown that especially reports about the suicide of famous people or those in highly sensational language are problematic. “Especially when a high volume of posts is made on social media in the context of sensational suicides, there is a risk that vulnerable people will slip deeper into the crisis as a result and that there will be an increase in suicides,” Niederkrotenthaler continues to call for a responsible approach to the topic.
HARDLY ANY STUDIES ON SOCIAL MEDIA
Although the influence of social media has increased strongly in recent years, especially in dealing with suicidal crises, there is currently hardly any research with meaningful data on this. Similar studies of traditional media usually work with relatively small data sets. This is due to the fact that the effort to manually categorise large data sets is enormous. New models can change that. “Our study shows that modern machine learning models can very precisely distinguish between different types of tweets that are relevant for prevention,” explains Metzler.
SEVEN MILLION TWEETS ANALYSED
The research team analysed a total of 7,159,610 million tweets – a quantity that would be almost impossible to handle manually. First, they optimised the model with a training set of 3,200 tweets. Only when it correctly recognises tweets relevant to suicide prevention in a completely new data set can such a model be considered reliable. “The models recorded many prevention tweets when NGOs actually do a lot of prevention work – for example on World Suicide Prevention Day or around Christmas and New Year’s Day,” explains Metzler, which underlines the reliability of the machine learning model.
HUGH POTENTIAL OF DEEP LEARNING
While older machine learning models only track the frequency of certain words, deep learning models can do much more complex analyses. For example, they take into account where a word is located in a sentence, as well as what comes before and after it. “Since words can have very different meanings in certain contexts, such models make our research more precise and, as in this case, can provide an important contribution to suicide prevention,” says Metzler.
“Our work is the first large-scale study to suggest that the daily volume of specific suicide prevention-related tweets is associated with higher daily rates of help-seeking behaviour and lower daily rates of suicide deaths,” Niederkrotenthaler summarises the study results.
The study “Association of 7 million+ tweets featuring suicide-related content with daily calls to the Suicide Prevention Lifeline and with suicides, United States, 2016–2018” was published in Australian & New Zealand Journal of Psychiatry and the study “Detecting Potentially Harmful and Protective Suicide-related Content on Twitter: Machine Learning Approach” was published in Journal of Medical Internet Research.