New insights are now available into the Twitter networks of the Islamic State and those who oppose them, thanks to a study by the RAND Corp.
Efforts to thwart the group on the social media site have made headlines in recent months, but researchers are also interested in filling the vacuum created by suspended pro-ISIS accounts by using targeted messaging that focuses on avoiding radicalization in the first place.
Since the San Francisco-based company first began banning accounts of ISIS supporters in 2015, more than 360,000 accounts have been removed from the microblogging site.
In the study U.S. Social Media Strategy Can Weaken ISIS Influence on Twitter, researchers used advanced network and lexical analysis to look at more than 23 million Arabic language tweets to determine their support or opposition for ISIS, who they are and what they are saying, and how they are connected across the larger network.
RAND Corp. engineer Elizabeth Bodine-Baron explained that the process was more than simply categorizing those who supported or opposed the radical militant group. The study also required a deep dive into the respective communities of the people on either end of the conversation.
By looking at the differences in how account holders referred to the West or ISIS, analysts were able to separate them into pro- and anti-ISIS categories with considerable accuracy.
“We had anecdotal evidence that people who were opposed to the Islamic State would use the term ‘Daesh,’ and people who were pro would use the term ‘Islamic State,’” she said. “By separating out whether you are using primarily the term 'Daesh' or primarily 'Islamic State,' we were actually able to quickly identify an account as pro-ISIS or anti-ISIS.”
The moniker Daesh is unfavorable to the group because of its similarity to the Arabic word for “to crush or trample,” but it can also mean “bigot,” according to an NBC News report.
“In aggregate, by comparing these two buckets; the people using Daesh and the people using Islamic State, we could definitely say that in aggregate, if you are using one term predominantly, you are going to be opposed or pro,” Bodine-Baron said.
What began as the identification of some 20,000 different communities was honed down, through the advanced networking algorithms, to what researchers identify as four main metacommunities: Shia, Syrian mujahideen, ISIS supporters and Sunni.
The team was then able to determine how the individuals involved in the conversation were connected by examining not only the position various account holders took on ISIS, but also the “edges” that connect the networks and the key issues participants care about.
This piece of the research is important, Bodine-Baron said, because it could ultimately allow for the development of more targeted counter-ISIS messaging that could help to quell the recruitment of supporters.
“The issue is you can’t just design a one-size-fits-all message for everyone," she said. "It has to be tailored to the different concerns of the different populations that could potentially be at risk for radicalization, and furthermore, it needs to be coming from credible Muslim voices, possibly even regional Muslim voices."
As for the removal of active extremist accounts, the research team said the effort to date seems to have had an effect on the overall social media campaign.
“We really want to see the account suspension campaign continue because that denies them their platform being able to just spread their propaganda without hindrance," Bodine-Baron said. "It forces them into less public spaces.”
In examining the dataset, Bodine-Baron said supporters were initially outnumbered 6-to-1, which would grow to nearly 30-to-1 by the end of the research period. She said the shrinking support for the group on Twitter was likely due to the aggressive deletion of associated accounts, but she qualified that more research would be needed to confirm that hypothesis.
In the larger network's analytics environment, this research is significant in that it could extend into other areas outside of examining polarizing organizations. It proves that big data sets can be analyzed quickly and effectively.
“One of the coolest things about what we’ve done is essentially do this proof of concept of where can these automated techniques for analyzing a really big dataset, where can they really help?” Bodine-Baron said. “By using these tools and specifically combining the network analysis and lexical analysis, we get these really powerful results where we can say, 'OK, here is this group of users and here is what they care about,' and being able to more or less automatically … figure out that these are the key themes we should be focusing our counter-messaging effort on.”