Abstract Objectives Social media platforms such as Facebook are used by both radicals and the security services that keep them under surveillance. However, only a small percentage of radicals go on to become terrorists and there is a worrying lack of evidence as to what types of online behaviors may differentiate terrorists from non-violent radicals. Most of the research to date uses text-based analysis to identify "radicals" only. In this study we sought to identify new social-media level behavioral metrics upon which it is possible to differentiate terrorists from non-violent radicals. Methods: Drawing on an established theoretical framework, Social Learning Theory, this study used a matched case-control design to compare the Facebook activities and interactions of 48 Palestinian terrorists in the 100 days prior to their attack with a 2:1 control group. Conditional-likelihood logistic regression was used to identify precise estimates, and a series of binomial logistic regression models were used to identify how well the variables classified between the groups. Findings: Variables from each of the social learning domains of differential associations, definitions, differential reinforcement, and imitation were found to be significant predictors of being a terrorist compared to a nonviolent radical. Models including these factors had a relatively high classification rate, and significantly reduced error over base-rate classification. Conclusions Behavioral level metrics derived from social learning theory should be considered as metrics upon which it may be possible to differentiate between terrorists and non-violent radicals based on their social media profiles. These metrics may also serve to support textbased analysis and vice versa.

中文翻译：

---

激进主义的面孔：通过社交媒体资料区分暴力和非暴力激进分子

摘要 目标 Facebook 等社交媒体平台被激进分子和保护他们监视的安全服务所使用。然而，只有一小部分激进分子会继续成为恐怖分子，而且令人担忧的是，缺乏证据表明哪些类型的网络行为可以将恐怖分子与非暴力激进分子区分开来。迄今为止的大多数研究都使用基于文本的分析来识别“激进分子”。在这项研究中，我们试图确定新的社交媒体级别的行为指标，可以根据这些指标将恐怖分子与非暴力激进分子区分开来。方法：借鉴既定的理论框架，社会学习理论，本研究使用匹配的病例对照设计，将 48 名巴勒斯坦恐怖分子在袭击前 100 天内的 Facebook 活动和互动与 2:1 对照组进行比较。条件似然逻辑回归用于确定精确估计，一系列二项逻辑回归模型用于确定变量在组之间的分类情况。发现：与非暴力激进分子相比，来自每个社会学习领域的差异关联、定义、差异强化和模仿的变量被发现是恐怖分子的重要预测因素。包含这些因素的模型具有相对较高的分类率，并且比基率分类显着减少了错误。结论 源自社会学习理论的行为水平指标应被视为可以根据其社交媒体资料区分恐怖分子和非暴力激进分子的指标。这些度量还可用于支持基于文本的分析，反之亦然。