Twitter posts can predict if protest will become violent

Posts on social media sites like Twitter can predict whether a protest is likely to turn violent, a study has found. People are more likely to endorse violence when they moralise the issue that they are protesting, and when they believe that others in their social network moralise that issue too, researchers said. "Extreme movements can emerge through social networks. We have seen several examples in recent years, such as the protests in Baltimore and Charlottesville, where people's perceptions are influenced by the activity in their social networks," said Morteza Dehghani, a researcher at University of Southern California in the US.

"People identify others who share their beliefs and interpret this as consensus. In these studies, we show that this can have potentially dangerous consequences," said Dehghani. Utilising a deep neural network - an advanced machine learning technique - to detect moralized language, the scientists analysed 18 million tweets posted during the 2015 Baltimore protests for Gray, 25, who died as police took him to jail.

The study, published in the journal Nature Human Behavior, investigated the association between moral tweets and arrest rates, a proxy for violence. This analysis showed that the number of hourly arrests made during the protests was associated with the number of moralised tweets posted in previous hours. In fact, tweets containing moral rhetoric nearly doubled on days when clashes among protesters and police became violent.

Social media sites such as Twitter have become a significant platform for activism and a source for data on human behaviours, which is why scientists utilize them for research. Recent examples of movements tied to social media include the #marchforourlives effort to seek gun control, the #metoo movement against sexual assault and harassment, and #blacklivesmatter, a campaign against systematic racism which began in 2014 after the police-involved shooting death of Michael Brown, 19, in Ferguson, Mo.

A more violent example is the Arab Spring revolution, which began in Tunisia in late 2010, and set off protests in many other countries, including Egypt and Libya, that forced changes in their leadership.
In Syria, clashes escalated into a war that has killed hundreds of thousands of people and displaced a multitude of refugees.

The scientists developed a model for detecting moralised language based on a prior, deep learning framework that can reliably identify text that evokes moral concerns associated with different types of moral values and their opposites.

"Social media data help us illuminate real-world social dynamics and test hypotheses in situ. However, as with all observational data, it can be difficult to establish the statistical and experimental control that is necessary for drawing reliable conclusions," said Joe Hoover, from University of Southern California in the US.

To make up for this, scientists conducted a series of controlled behavioural studies, each with more than 200 people, how much they agreed or disagreed with statements about the use of violence against far-right protesters after they had read a paragraph about the 2017 Charlottesville, Va., clashes over the removal of Confederate monuments.

The more certain people were that many others in their network shared their views, the more willing they were to consider the use of violence against their perceived opponents, the scientists found.