CAREER: Social Response-Powered Misinformation Detection, Robustness, and Correction

该项目将发明检测和纠正在线平台上错误信息的方法。网络错误信息对公共卫生、民主、科学和社会构成了令人担忧的威胁。大规模处理错误信息仍然是一项紧迫的挑战，因为目前的解决方案依赖于专业事实检查员或版主的有限资源，这些资源既不能扩展到新出现的信息问题，也不能直接解决如何应对现场的错误信息。该项目将通过开发强大的检测模型来应对这些挑战，该模型利用用户对社交媒体帖子的回复来识别潜在的不可信信息。该团队还将设计一个反回应生成工具，帮助日常用户有效地回应错误信息，利用开发的模型以及现有的事实核查资源和沟通中的最佳实践，建议对错误帖子的可能回应，帮助读者评估这些帖子。总之，拟议的工作将提高社会的信息素养，减少接触错误信息的人数。该团队还将开发跨学科的课程和研究机会，这将扩大学生解决社交媒体系统中错误信息的工具包，并扩大参与其中的学生的范围。该项目将推进错误信息、图神经网络、对抗性学习和社会网络分析方面的科学知识。一般的方法是利用普通用户对在线帖子的社会反应，如支持、质疑、不相信或反对主张，以强有力地发现错误信息并提出纠正措施。围绕检测，该项目将开发新的签名动态图神经网络模型和网络增强方法来解决网络稀疏性问题。围绕健壮性，该项目将创建对对抗性操作具有健壮性的检测模型，通过更好地模拟攻击者群体进行的对抗性攻击，然后创建防御措施，优化对社交媒体评论的虚假响应注入。在纠正方面，该项目旨在通过开发文本生成方法，对估计含有信息的帖子提出有效的回应建议，从而增强社交媒体用户纠正错误信息的能力；这些方法将根据从专业事实核查组织收集的数据进行培训，并在一系列研究中进行评估。该项目还将产生针对错误信息的新模型、数据集、基准和工具，这将促进这些主题的未来研究。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 project will invent methods to detect and correct misinformation on online platforms. Online misinformation poses an alarming threat to public health, democracy, science, and society. Addressing misinformation at scale remains a pressing challenge as current solutions rely on the limited resources of professional fact-checkers or moderators, which neither scales to newly emerging information issues nor directly addresses how to respond to misinformation in situ. This project will address these challenges through developing robust detection models that leverage user-generated responses to social media posts to identify potentially non-credible information. The team will also design a counter-response generation tool that can help everyday users effectively respond to misinformation, leveraging the models developed along with existing fact-checking resources and best practices in communication to suggest possible responses to incorrect posts that will help readers assess them. Together, the proposed work will boost information literacy in society and reduce the number of people exposed to misinformation. The team will also develop interdisciplinary coursework and research opportunities that will broaden both students’ toolkits for addressing misinformation in social media systems and the range of students who engage in it.This project will advance scientific knowledge in misinformation, graph neural networks, adversarial learning, and social network analysis. The general approach is to leverage the social responses that ordinary users make on online posts, such as supporting, questioning, disbelieving, or countering claims, to robustly detect misinformation and suggest corrective responses. Around detection, the project will develop novel signed dynamic graph neural network models and network augmentation methods to address network sparsity issues. Around robustness, the project will create detection models that are robust to adversarial manipulation, by better modeling adversarial attacks carried out by groups of attackers, then creating defenses that optimize against fake response injections into social media comments. Around correction, the project seeks to empower social media users to correct misinformation by developing text generation methods to suggest effective counter-responses to posts estimated to contain information; these methods will be trained on data collected from professional fact-checking organizations and assessed in a series of studies. The project will also result in new models, datasets, benchmarks, and tools around misinformation that will promote future research on these topics.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.