Collaborative Research: SaTC: CORE: Small: Detecting Accounts Involved in Influence Campaigns on Social Media

在线社交网络的日益普及为恶意行为者传播虚假信息打开了大门。敌对的民族国家正在美国精心策划虚假信息运动，目的是影响公共话语或推动有利于特定国家议程的谈话要点。这些活动通常是在被称为“巨魔”的虚假社交媒体账户的帮助下进行的，这些账户扮演虚构的角色。这些喷子彼此互动，并与毫无戒心的社交网络用户互动，目的是使讨论两极分化，并进一步虚假叙述。尽管虚假信息对我们的社会构成了严重威胁，但社交网络仍在努力应对这种威胁，采取了事后删除恶意账户的反应性措施。该项目的目标是开发能够自动识别社交网络上的喷子账户的技术，以便更快地将其删除。为了开展这个项目，调查人员正在收集有关Twitter和Reddit确定的巨魔账户的信息，这些账户属于几个外国的虚假信息活动。这些数据被用来训练机器学习算法，这些算法能够自动识别Twitter和Reddit上的其他巨魔账户。该项目遵循两个并行的研究任务。在第一个任务中，研究团队旨在训练监督学习算法，以学习已知巨魔账户的典型行为，目标是识别更多行为相似的账户，从而成为同一虚假信息活动的一部分。初步结果显示，喷子账户表现出在真实账户中不常见的特殊互动模式。在第二个任务中，研究人员将致力于识别喷子账户的典型特征，而不管它们属于哪个活动，并使用迁移学习技术来识别新出现的虚假信息活动。学生们参与了研究的各个阶段。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。

The growing popularity of online social networks has opened the door to false information being disseminated by malicious actors. Hostile nation states are orchestrating disinformation campaigns in the United States with the goal of influencing public discourse or pushing talking points that favor a particular country’s agenda. These campaigns are typically carried out with the help of fake social media accounts known as trolls, which pose as fictitious personas. These trolls interact with each other and with unsuspecting social network users with the goal of polarizing discussion and furthering false narratives. Despite the serious threat that disinformation poses to our society, social networks are struggling to keep up with the threat, resorting to reactive measures that remove malicious accounts after the fact. The goal of this project is to develop techniques able to automatically identify troll accounts on social networks so that they can be more swiftly removed.To carry out this project, the investigators are collecting information about the troll accounts identified by Twitter and Reddit as belonging to several foreign countries’ disinformation campaigns. These data are used to train machine learning algorithms able to automatically identify other troll accounts on Twitter and Reddit. The project follows two parallel research tasks. In the first task, the research team aims to train supervised learning algorithms to learn the typical behavior of known troll accounts, with the goal of identifying more accounts that behave similarly and are therefore part of the same disinformation campaign. Preliminary results show that troll accounts exhibit peculiar interaction patterns that are uncommon in real accounts. In the second task, the researchers will work towards identifying traits that are typical of troll accounts regardless of the campaign that they belong to, and use transfer learning techniques to identify emerging disinformation campaigns. Students are involved at all stages of the research.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.