Check out this new article by Gillian Kant, Talip Alkhayer, Timo Kivimäki, and Christoph Weisser, titled “The Word and the Bullet: Out-Grouping and Threat Framing as Predictors of Islamic State Targeting, 2015–2020.” The study digs into whether what the Islamic State says in its weekly newsletter, Al-Naba, can predict who it attacks next.
The study sets out to test whether shifts in ISIS’s language about its enemies can forecast its targeting behaviour. The authors ask: when ISIS rhetoric frames certain groups as threatening or dangerous, does that predict attacks on those groups in the months that follow? The work goes beyond describing long-term “enemy types” to explain short-term variation, that is, who gets targeted when, and why.
They build on Social Identity Theory and Integrated Threat Theory, arguing that terrorist violence is more often driven by fear of outgroups than by hatred or religious difference. Terrorists act when they perceive another group as both identifiable and threatening enough to justify risk.
The research combines Arabic-language text analysis with terrorism event data. Using all 272 issues of Al-Naba from October 2015 to December 2020. The used topic modelling (Latent Dirichlet Allocation) to extract recurring themes from ISIS rhetoric such as references to religious difference, hatred, unreachable enemies, or explicit threats.
The authors focused on pages mentioning outgroups such as Shia, Alawites, Kurds, Christians, Westerners, and regional “near enemies”, and cleaned and tokenised the Arabic text. They then compared topic prominence in each month’s newsletter with ISIS attack data from the Global Terrorism Database (GTD), covering more than 4,500 attacks and 23,000 fatalities in Iraq and Syria.
They used Spearman correlations and Granger causality tests to see if rhetorical themes predicted subsequent targeting patterns (and vice versa). The tests were run with a six-month lag to capture realistic temporal relationships between speech and action.
The analysis revealed five recurring topics in ISIS rhetoric. Of these, only one, threat-based framing of outgroups, strongly predicted actual targeting. When Al-Naba emphasised that specific groups posed an existential threat to ISIS, those same groups were far more likely to be attacked soon after.
In short, fear and threat, not hatred or ideology, were the most reliable predictors of who ISIS struck next.
For policymakers, the findings offer a new way to anticipate terrorist targeting: by tracking shifts in threat-based rhetoric within extremist communications. Monitoring changes in how groups are framed (as threatening, reachable, or existentially dangerous) may provide early warning signals of attack risk.
For researchers, the study showcases the power of Arabic-language natural language processing to illuminate extremist decision-making. Perhaps most importantly, the results challenge the common belief that extremist violence is driven by hatred alone. Fear, framed through language that constructs dangerous “outgroups,” appears to be the more powerful motivator, and understanding that distinction could reshape how we study and prevent terrorism.