Increased usage of social media led to formation of online communities of terrorist groups for discussing many violent plans. These online communities present big data to researchers to identify hidden patterns and behaviors to generate actionable intelligence which can be useful for security agencies. Recruiting new people over online social media is on emerging trend which presents how internet is exploited by terrorist groups. In this paper, we present automatic model to detect online cyber recruitment over social media. Online discussions regarding recruitment of terrorists or building new connections for recruitment of violent extremist have been labelled. Two experts labelled 730 messages from five dark web discussion forums named ‘Ansar1’, ‘Gawaher’, ‘Islamic Awakening’, ‘Islamic Network’ and ‘Mywic’ as YES (recruitment) and NO (non-recruitment). Statistical analysis has been done on two independent labelled messages to find mutual agreement between two judges. Kohen’s Kappa coefficient computed is 0.87 at p = 0.01 which signifies higher mutual agreement. Five machine learning classifiers namely support vector machine, logic boosting, random forest, generalized linear model and maximum entropy based model are developed to further classify recruitment labels. To the best of our knowledge no such work has been done by collaborating data from multiple dark web discussion forums to automatic identify online recruitment of terrorists. Our proposed model presents smart solution with usage of computational techniques to quantify terrorist behavior analysis and detect online recruitment of violent extremists over online social media and dark web forums.

社交媒体使用的增加导致恐怖分子网上社区的形成，以讨论许多暴力计划。这些在线社区向研究人员提供大数据，以识别隐藏的模式和行为，以生成可操作的情报，这些情报对安全机构很有用。通过在线社交媒体招募新人正处于新兴趋势，这表明恐怖组织如何利用互联网。在本文中，我们提出了一种自动模型来检测社交媒体上的在线网络招聘。关于招募恐怖分子或为招募暴力极端主义建立新联系的在线讨论已被标记。两名专家标记了来自五个暗网讨论论坛的730条消息，这些论坛名为“ Ansar1”，“ Gawaher”，“伊斯兰觉醒”，“伊斯兰网络”和“ Mywic”分别为“是”（招募）和“否”（非招募）。已经对两条独立标记的消息进行了统计分析，以找到两位法官之间的共识。Kohen的Kappa系数在p = 0.01时为0.87，这表明更高的相互一致性。开发了五个机器学习分类器，即支持向量机，逻辑提升，随机森林，广义线性模型和基于最大熵的模型，以进一步对招聘标签进行分类。据我们所知，还没有通过协作来自多个暗网讨论论坛的数据来自动识别恐怖分子的在线招募工作而进行过此类工作。