HCC: Small: Assessing Radicalization and Deradicalization Online

这项研究将为衡量网络社区的激进化，更重要的是去激进化奠定基础，方法是追踪个人长期大规模的行为趋势，并识别那些脱离了亚文化的成员的模式，他们曾在激进化过程中加入亚文化，但已经开始去激进化。网络极端主义和激进化已经成为人们日益关注和活跃的研究领域。有广泛的跨学科工作研究在线极端主义团体，但基于计算和数据科学的去激进化过程研究有限。主要的实证研究将分析最近收集的超过800万在线论坛帖子的大型语料库，包括与一个极端主义亚文化相关的五个流行论坛的综合档案。这项工作的实际意义有可能有利于社会福祉，无论是在线还是离线，并加强国家安全。这项工作包括五个核心项目：(1)检测激进化将建立亚文化语言的正式词汇，这是群体归属感的信号，并建立一套工具来检测成员之间的仇恨言论。(2)“检测激进化”将分析每个用户的语言使用趋势，发现模式，并进行内容分析，识别激进化的标记。(3)社会激励将研究社会互动及其与激进化的关系，以了解将社会激励与真实信念联系起来的理论的适用性。(4)检测去极端化将识别那些脱离论坛的成员和那些转向试图离开社区的成员的支持小组的成员的去极端化标志。(5)分类去极端化将建立机器学习模型，自动识别可能考虑去极端化的成员。本研究开发的模型和词汇将为更广泛地理解极端主义团体和亚文化的社会和规范动态提供方法。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

This research will develop a foundation for measuring radicalization and, importantly, deradicalization, in online communities by tracking individual behavioral trends over time and at scale, and identifying patterns among members who disengage from the subculture they had earlier joined in a process of radicalization, but had begun to deradicalize. Online extremism and radicalization has been a growing concern and active area of research. There is a broad body of cross-disciplinary work studying online extremist groups but limited computational and data science-based research on the deradicalization process. The primary empirical research will analyze a recently-collected large corpus of over 8 million online forum posts, including comprehensive archives of five popular forums related to one extremist subculture. The practical implications of this work have potential to benefit the well-being of society, both online and offline, and to bolster national security.This work includes five core projects: (1) Detecting Radicalization will build a formal lexicon of subcultural language, a signal of belonging in the group, and build a suite of tools to detect hate speech among members. (2) Detecting Radicalization will analyze trends of language use per user to find patterns and conduct content analysis to identify markers of radicalization. (3) Social Incentives will study social interaction and its relationship with radicalization to understand the applicability of theories that connect social incentives and genuine beliefs. (4) Detecting Deradicalization will identify markers of deradicalization among members who disengage from forums and those who turn to support groups for members who are trying to leave the community. (5) Classifying Deradicalization will build machine-learning models to automatically identify members who may be considering deradicalization. The models and lexicons developed in this research will provide methods for understanding the social and normative dynamics of extremist groups and subcultures more broadly.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.